Experimental assessment with data-driven machine learning-based prediction of compressive strength of waste natural fiber-reinforced sustainable concrete Authors: Prince Paul, Md Kawsarul Islam Kabbo, Rajarshi Das Gupta, Md Habibur Rahman Sobuz, Prosanjit Baral, Mohammed Jameel, Sani Aliyu Abubakar Published: 2025 ABSTRACT AbstractThis study presents a hybrid approach combining experimental evaluation and machine learning modeling for the compressive strength (CS) estimation of natural fiber-reinforced concrete (NFRC), which utilizes jute, coir, and bamboo fibers. A dataset of 444 concrete mix designs was compiled from the literature using seven input variables. Six advanced machine learning (ML) algorithms: Light Gradient Boosting Machine (LGBM), Extreme Gradient Boosting (XGB), Random Forest, Gradient Boosting, Decision Tree, and K-Nearest Neighbors were trained and tested with proper hyperparameter optimization. Among these, the LGBM model demonstrated superior accuracy in prediction with an R2 of 0.8637 and the lowest root mean square error of 4.19 MPa on the testing sets. To ensure interpretability and mix design optimization, SHapley Additive Explanations (SHAP) and Partial Dependence Plot (PDP) analyses were incorporated. The dominant predictors of CS were found to be cement content, water, coarse aggregate, and supplementary cementitious materials, while the curing period and fiber content also showed a small but meaningful effect. To further validate the modeling outcomes, experimental investigations were conducted by developing 10 mix combinations with varying percentages of coir fiber, which were then subjected to compressive strength and scanning electron microscopy tests. The outcomes also validated that the incorporation of natural fiber up to 0.75% gradually increased the CS to a maximum of 41.3% after 28 days of curing, while further addition reduced performance. The dual-approach-based (ML and experimental) outcomes could assist in the sustainable advancement of the infrastructure industry as a potential solution for cost-effective, large-scale production of NFRC.Keywords: Supervised machine learning, Natural fiber, Microstructural assessment, Compressive strength, Parametric optimizationSubject terms: Engineering, Materials science, Mathematics and computing Introduction Due to its capabilities and being easy to use, concrete continues to be one of the most popular materials globally, playing an essential role in the building of residential infrastructure and cities1. With rising global demands, experts are predicting that the world will produce 5.5 billion tons of concrete in a year by 20502. Such considerable usage not only demonstrates the adaptability and durability of concrete but also the importance of seeking innovative and sustainable production and improved material performance3. Concrete is inherently fragile because of its brittle tensile strength and strain capacity4,5. Moreover, studies have shown that concrete suffers from low tensile strength, high porosity, cracking sensitivity, minimal impact resistance, water permeability, and low chemical durability6,7. To address this issue, several studies have shown that concrete includes fibers such as steel, carbon, jute, coir, palm, and cellulose to improve its efficiency as a stabilizing and reinforcing element8–11. Several studies reveal that natural fibers are a more sustainable alternative to synthetic fibers, improving the compressive and tensile strength, crack resistance, fracture tolerance, and stress adaptability of concrete12–15. Past research has demonstrated that cellulose fiber-reinforced concrete increases strength, toughness, and durability, with the optimal content being 0.9 kg/m³ in extreme environmental conditions16. A study of bamboo fiber-reinforced concrete indicates enhanced tensile, flexural, impact, and ductile properties, with an optimum fiber content of 1.2% by weight of cement17. One of the studies shows that the incorporation of the 0.25% jute fiber improves compressive and flexural strength18. An investigation revealed that 1% Kenaf fiber reinforcement in high-strength concrete composites enhances tensile strength and toughness; however, compressive strength is reduced19. Experts agree that sometimes natural fibers, though useful in concrete reinforcement, reduce concrete strength due to a lack of hydration and fewer pozzolanic reactions20,21. To resolve this problem, researchers have employed Supplementary Cementitious Materials (SCM) like fly ash, slag, silica fume, and metakaolin to enhance durability, porosity, and fiber-matrix bond in the fiber-reinforced concrete22–26. Al-Naghi et al.27 revealed that the addition of wheat straw ash at 10%, silica fume at 7%, and coconut fiber at 1.5% in a recycled aggregate concrete produced 30.7 MPa of compressive strength and 3.89 MPa of tensile strength. Subedi et al.28 showed that the addition of 1% Himalayan Giant Nettle fiber with SCM, such as 20% fly ash and 20% rice husk ash, enhanced compressive (16.2%), flexural (33.3%), and tensile strength (36.9%). The fibers have an exceptionally good ability to enhance strength by reducing the development of cracks and evenly distributing stresses in the concrete29. To understand the mechanical behavior of different fiber-reinforced concrete incorporating different SCM, robust studies and tests are needed. But concrete production has environmental consequences because it releases considerable quantities of carbon dioxide (CO2) into the atmosphere, which causes climate change24,30,31. These limitations justify the necessity of more adaptable and robust test techniques in sustainable concrete systems. In recent years, machine learning (ML) algorithms have been used to solve these limitations by applying them to predictive modeling of structural systems32–37. Along with ML modeling, numerical simulation also has a significant application to reduce the time-consuming laboratory trials38–40. Pal et al.41 revealed that K-Nearest Neighbors (KNN) forecasted the tensile strength of fiber-reinforced rubberized recycled aggregate concrete (FR3C) with an R2 of 0.944, and it was shown that various synthetic fiber content, W/C ratio, and rubber content played a critical role in determining the mechanical properties. Manan et al.42 developed an AI-based model to predict axial load-deflection behavior of recycled concrete powder and steel fiber reinforced columns with high accuracy (R2 = 0.95), using 196 datasets. Ray et al.43 evaluated the effectiveness of Support Vector Machine (SVM) and Gradient Boosting Machine (GBM) to predict the compressive and splitting tensile strength of ceramic waste-based and nylon fiber-based concrete on a dataset of 162 records, and GBM (R2 = 0.981) achieved higher results than SVM. Also, considerable research has been conducted on developing supervised ML models to predict the individual effect of either SCM with synthetic fiber, including marble dust (MD) with glass fiber (GF)44, Volcanic Pumice Powder (VPP) with Nylon Fiber (NyF)45, ground granulated blast furnace slag (GGBS) with glass fiber (GF)46, hybrid fiber such as steel with polyvinyl alcohol fiber (PVAF)47, and silica fume (SF) with polypropylene fiber (PPF)48, on mechanical properties. Few studies have been done about ML applications on natural fibers and SCM49–51. For instance, Kashyap et al.52 employed several ML models, such as Random Forest (RF), Artificial Neural Network (ANN), and SVM, to predict the strength of coconut fiber-reinforced concrete, where the RF model (R2 = 0.9841) outperformed all models. A study by Abdrahman et al.53 employed Gaussian Progress (GP), Linear Regression (LR), ANN, and SVM to predict the strength of sisal fiber-reinforced concrete composites, with ANN showing higher accuracy than the other models. Cakiroglu et al.54 developed predictive models, including Extreme Gradient Boosting (XGB) and Light Gradient Boosting Machine (LGBM) for tensile strength of natural-fiber-reinforced recycled aggregate concrete, XGB achieving the highest R2 = 0.95. Table 1 shows various AI techniques applied to predict the properties of SCM and fiber-reinforced concrete, showing that most studies relied on single ML models. Although previous research has evaluated the effect of fiber and SCM materials in concrete and the use of machine learning in predicting material performance, there remains a limitation in understanding the combined impacts of natural fibers and SCM materials on structural components. Additionally, the use of machine learning to optimize their mix design remains challenging due to the non-linear nature of the dataset and the need for large datasets. Table 1Previous ML studies on predicting concrete strength.MaterialsModeling techniquesData rangeReferencesFA + Virous fiberRF, KNN, SVM, DT, ANN, AdaBoost, GB, CB, and XGB.34641RCP + StFRF19642CCA + NyFSVM and GBM16243MD + GFXGB, RF, and SVM18144VPP + NyFRF, KNN, RF-KNN17445GGBS + GFANN, XGB, RF, SVM, and GB28546StF + PVAFXGB, DT, RF, SVM19247SF + PPFLR, ANN33648RCPANN, DTR, RF, SVR, XGB, GB, KNN, LR, ENR, and RR.27049CFRF, SVM, and ANN25652SiFSVM, GP, and ANN26753RCA + NFXGB, LGBM, and ETR38254FA = fly ash, RCP = recycled concrete powder, StF = steel fiber, CCA = coarse ceramic aggregate, CF = coconut fiber, SiF = sisal fiber, RCA = recycled coarse aggregate, NF = natural fiber, DT = decision tree, CB = CatBoost, RR = Ridge regression, ENR = elastic net regression, ETR = extra tree regressor. Previous ML studies on predicting concrete strength. FA = fly ash, RCP = recycled concrete powder, StF = steel fiber, CCA = coarse ceramic aggregate, CF = coconut fiber, SiF = sisal fiber, RCA = recycled coarse aggregate, NF = natural fiber, DT = decision tree, CB = CatBoost, RR = Ridge regression, ENR = elastic net regression, ETR = extra tree regressor. To address these limitations, the current study proposes six supervised ML-based models to predict the compressive strengths of natural fiber-reinforced concrete (NFRC) incorporating supplementary cementitious materials (SCM), as well as investigating their performance through comparative evaluation. Here, jute, bamboo, and coir fibers are utilized as natural fibers, and different SCMs of similar properties are considered for the natural fiber-reinforced concrete (NFRC). The proposed study enhances the efficiency and sustainability of concrete research and development by reducing the necessity for trial-and-error mix designs. In this research, a significantly large dataset of 444 data points acquired from previous literature is utilized, and seven key input parameters are employed. Furthermore, the incorporation of SHAP (SHapley Additive explanations) and Partial Dependence Plots (PDP) facilitated an extensive examination of parameter impact, providing critical insights for optimizing mixture ratios. These insights can inform the design of concrete mixes that attain the desired compressive strength with fewer experimental trials. Research methodology Machine learning algorithms In this investigation, six models (RF, GB, XGB, LGBM, DT, and KNN) were employed due to their ability to handle complex nonlinear data interactions effectively while providing a strong balance among interpretability and prediction accuracy. These models offer high efficiency and robustness for tabular data with limited samples and serve as a clear baseline to demonstrate the impact of variables. In contrast, hybrid machine learning approaches require many computational resources and a robust dataset, which are not always feasible in practical field conditions. Random forest (RF) regressor In the world of tree-based ensemble models, Random Forest is a widely favored method that’s frequently applied in many data analysis and machine learning tasks. The Random Forest algorithm works by using a collection of decision trees that have been trained using the bagging approach55. The model trains several randomly created decision trees at the same time, allowing for parallel processing. The technique provides configuration options to fine-tune the design, scale, and randomness level of every individual tree. RF improves the accuracy of the model because it chooses a few of the features in each tree, and aids in preventing overfitting. A good feature is selected at each node of the decision tree in such a manner that the data is separated into the left and right descendant subtrees. This is critical in the enhancement of the generalization ability of the model. The training process of RF happens in two main stages: first, the decision trees are built, and then their predictions are combined. Bootstrap sampling is used to generate random, same-sized subsets of the data for training each tree individually. The next stage involves combining the outputs of the individual trees using bagging. The result is produced by performing an average of the results of several decision trees trained on different subsets of the data, as per the equation given below 1. Each of the decision trees, Yb, is trained based on distinct sets of the unspecified data set instances . In this equation, B denotes the total lineups of decision trees that are incorporated in the model56. Gradient boosting (GB) regressor Gradient Boosting is a frequent type of forward-stage ensemble learning, frequently applied in machine learning to both classification and regression problems57. This method builds a powerful model by integrating multiple decision trees, known as GB regression trees. Unlike random trees, GB regression trees are not generated randomly; instead, an alternative approach they adopt is strict dictionary pruning. Similar to AdaBoost, the GB model adds predictors to the aggregate sequentially, with each successive predictor trying to resolve the mistakes of the preceding predictors. However, unlike AdaBoost, GB learns from the residual errors left by the prior predictors, refining its model based on these mistakes. The method of the GB model has been found to combine several weak learners (usually regression trees) into one strong model. It trains step by step, each time focusing on the errors made before, so predictions keep getting better. Equation (2) shows the algorithm train additive in the GB step by step. The GB training output incorporates the regression decision trees as its weak learners, which are represented by Gm(x). The outputs of the weak learners are focused into the algorithm, and each of the iterations adds one more estimator to the model. The effect of each decision tree on the overall prediction is regulated by a parameter termed the learning rate to avoid the model overfitting, as indicated in Eq. (3). Extreme gradient boosting (XGB) model The XGBoost algorithm is an optimized version of the Gradient Boosting (GB) algorithm58. As a supervised machine learning method, XGBoost has attracted considerable interest and widespread use in recent times due to its high accuracy, efficiency, and scalability. XGB is built on the boosting framework, which works by consecutively adding weak learners to create a stronger overall model55. Through repeated iterations, XGB refines the model by learning from the errors of previous weak learners, ultimately producing a highly accurate prediction model (shown in Fig. 1). The XGBoost employs the second-order Taylor series of the loss function in order to achieve accuracy improvement in prediction, including the first and second derivatives in every iteration. The method shall be figuring out the negative gradient of the loss function, and updating the model sequentially to fix the model and reach the desired accuracy. Graphical representation of the XGB model. A major aspect of XGB is that it employs a control method to avoid overfitting. It has an adaptive term on the equation of loss, and this allows us to control the generalization capacity of the model by preventing excessive complexity. The prediction function of the XGBoost algorithm, which defines its phase of prediction, is represented in Eq. (4). Here, is the learner at phase , is the prediction at the phase is the prediction from the previous phase, and is the input data variable. The XGB model has a built-in way to measure how well a chosen model performs, helping to avoid overfitting without sacrificing data processing performance59. In which l is the LF, n is the number of observations, and is the regularization term in the model, which is calculated using the following equation below. where, denotes the score values assigned to each leaf, sets the minimum loss needed to justify splitting a leaf further, and controls regularization to keep the model from overfitting. Light gradient boosting machine (LGBM) model LGBM is a strong machine learning algorithm, which is an integration of decision trees and gradient boosting towards improved outcome of model prediction60,61. However, due to its high accuracy as well as capacity to scale with datasets of larger size via a histogram-based data discretization approach to continuous data, LGBM is particularly applicable in regression tasks. A key advantage of the LGBM algorithm is its rapidity in training and prediction. The technique is distinguished by its rapid training, elevated prediction accuracy, little memory use, and capacity for parallel processing, rendering it optimal for extensive datasets60,61. LGBM employs a leaf-wise growth strategy, where the data is split at the leaf level, focusing on the most significant data points, shown in Fig. 2. This allows the algorithm to construct trees more efficiently, improving model performance with each iteration. The model’s final predictions are determined by taking a weighted average of the outputs from each decision tree, boosting its overall performance and reliability. Visualization of the LGBM model. Here, stands for the dataset’s overall entropy, is the proportion of that falls into the class , refers to all the possible classes, is part of where a specific feature equals value , and is that feature’s value. Decision tree (DT) regressor A Decision Tree (DT) is a hierarchical model that works by applying a series of decision rules to split input variables into increasingly uniform groups systematically62. Its primary objective is to identify a group of rules that can easily predict an outcome based on specific input data. Depending on the nature of the target variable, DTs are classified as either regression trees (for continuous outputs) or classification trees (for categorical outputs). By deploying DTs as predictive and classificatory machine learning methods, many authors have been able to use DTs in a variety of real-world situations. One essential benefit of DT models is that they have an excellent ability to extract and model interactions among input variables. DTs have several advantages over the others, since they can process both quantitative and qualitative ordinal data without assuming a strong distribution of the data. In addition, DT models are easy to train and easy to interpret, which makes them attractive for applications requiring model interpretability. K-nearest neighbors (KNN) model The K-nearest neighbors (KNN) algorithm is generally a model of low efficiency in machine learning, although it retains all the points it views and afterwards classifies and estimates another point in such a way that the new point is the nearest to the training point63. The main purpose of this model is to predict a numerical result by finding the mean of the values of the k-nearest neighbors to the observed value. To compute nearest neighbors, KNN uses an inverse distance weighted average, where closer neighbors are weighted most. In regression tasks, KNN relies on three main distance functions to determine how far apart neighbors are. These functions include: where p represents the order between the X and Y points, and Ai and Bi stand for the ith dimensions of the A and B points. Hyperparameter tuning Hyperparameter tuning influences the performance of the algorithms greatly, and the choice of which hyperparameters would be the most useful is the key to creating any successful ML algorithms. The scikit-learn library used in this study has the GridSearchCV class, which was used to automatically tune the hyperparameters of these machine learning models. Hyperparameters are essential in the development of machine learning algorithms that are accurate. The use of gridSearchCV is an effective and trustworthy method of hyperparameter fine-tuning, but it is time-consuming. This paper has discussed the hyperparameters of machine learning algorithms applicable in the given research, as shown in Table 2. Table 2Hyperparameters for ML models used in this study.ModelKey parameter tunedGrid search rangeOptimum valuesRandom forestNumber of estimators[200, 500, 1000]1000Max depth[20, 30, 50]50Min samples split[2, 5, 10]2random_state[100, 200, 300]200Gradient boostingNumber of estimators[100, 200, 400]200Learning rate[0.01, 0.05, 0.1]0.05Max depth[3, 5, 10]5random_state[50, 70, 100]70XGBoostNumber of estimators[100, 200, 400]200Learning rate[0.05, 0.1, 0.3]0.1Max depth[5, 10, 15]5random_state[50, 70, 100]70LightGBMNumber of estimators[200, 300, 500]500Learning rate[0.01, 0.05, 0.1]0.05Num leaves[20,30, 40]30Max depth[5, 10, 15]5Boosting types[gbdt, rf, dart]gbdtDecision treeMax depth[5, 10, 15]10Min samples split[2, 5, 10]2Min samples leaf[1, 3, 5]5Criterion[squared_error, friedman_mse]squared_errorKNNNumber of neighbors[3, 5, 7]3Weights[‘uniform’, ‘distance’]uniformMetric[‘euclidean’, ‘manhattan’]euclidean Hyperparameters for ML models used in this study. SHapley additive explanations (SHAP) This model is an explanation method grounded in game theory, designed to evaluate predictions made by machine learning algorithms. It employs Shapley values, which are rooted in mathematical theory and originally developed for cooperative games, to determine the contribution of each participant (in this case, the input features) to the outcome, or “payout”, which is the model’s prediction60,64. The fundamental goal of SHAP is to evaluate the influence of each input variable on the model’s predicted outcome by observing how predictions change depending on the inclusion or removal of a feature from the model’s input set. This approach assigns a specific value to each feature, indicating its impact on the prediction. However, SHAP represents the model explanation as a combination of feature contributions, allowing for a transparent view of how each input affects the model’s behavior59,65–67. Here, Shapley values () show each feature’s impact, is a simplified input, and explains how the model makes predictions. Here, F denotes the complete set of input features, while S refers to a subset of F in which the feature at index i is excluded. Partial dependence plot (PDP) analysis It is a convenient instrument to visualize the interrelation of a series of input variables to the machine learning model output prediction. It may be helpful to demonstrate the variations of the model’s predictions as input parameters are varied (a more intuitive knowledge of how input parameters interact with the model)68. The PDPs are especially useful since they also reveal the average influence of each parameter on the predictions, which is needed to interpret complex models. PDPs are an interpretable method in a global scope, which reveals a global scenario of how each feature affects the predictions of the model58. With these plots, we can look at how the output depends on one of the features when all other features are fixed. For example, one-dimensional PDPs (PDPs-1D) can be used to evaluate the impact of a single feature on the predicted results69,70. This makes them especially useful for investigating straightforward, direct relationships between a feature and the model’s predicted value. In practice, PDPs can reveal non-linear relationships between predictors, interactions, or threshold effects that would otherwise remain hidden. Performance evaluation indicators Four common statistical indicators were carefully observed in the three phases of training, validation, and testing to measure how the prediction model performed. The mean absolute percentage error (MAPE), root mean square error (RMSE), the mean absolute error (MAE), and the determination coefficient (R2) were the measurements to check the performance of the prediction models. Based on these different measures, the accuracy and living value of the model could be well measured in a variety of aspects. The RMSE was specifically used to estimate the mean error of the observed and the predicted values of the observations. Another measure, which was used on the other hand, is the mean absolute percentage error (MAPE) to measure percentage errors of the prediction, and to inform about the overall accuracy of the model in relation to the actual values. Cumulatively, such measures offered a good baseline in order to determine the effectiveness of the model at various stages of testing and validation71–74. The following formulas describe how each performance accuracy is calculated. Here, represents the actual data points, is the predicted value, denotes the mean value, and n represents the total count of observations. Figure 3 depicts the complete workflow for developing and evaluating individual machine learning techniques aimed at predicting the compressive strength of natural fiber reinforced concrete (NFRC). Workflow of the construction of ML-based prediction of the CS of concrete. Results and discussion Analytical overview of dataset attributes In the context of the present investigation, a comprehensive dataset of about 444 data points was obtained from prior research papers. The dataset emphasizes concrete mix compositions and comprises critical input parameters, including water, natural fiber, supplementary cementitious materials (SCMs), sand, coarse aggregate, and curing period. The data sample for the concrete includes 444 samples of mix proportions, with 8 data attributes, among which compressive strength is regarded as the response variable. Additionally, an overview of the input and output datasets and their statistics can be found in Table 3. Each parameter’s statistical data is outlined in Table 3, accompanied by its minimum and maximum values, kurtosis, skewness, and standard deviation (std). Most values accumulate around the mean in cases of low standard deviation, such as for variables like Natural Fiber, SCM, and Compressive strength. However, the values get more dispersed as the std for Cement, Sand, and Coarse aggregate rises. Skewness, which can be zero, positive, negative, or undefined, assesses the imbalance in the probability distribution of a variable about the average value75. Positive skewness values suggest data density in Fig. 4e for Compressive strength and SCM. When skewness numbers are negative, it means that there is more data on the right side of Fig. 4a. Kurtosis measures the sharpness or flatness of a data distribution relative to the normal distribution. It helps assess whether the data has more or fewer extreme values, with typical values falling between − 10 and + 1076. Table 3 reports a higher kurtosis value for Curing Period (days), due to a small sample size. The presence of skewness and kurtosis in the dataset indicates non-normality in several input variables. This non-normality affects performance metrics (R², RMSE, MAPE, and MAE) and requires normalization techniques. However, since the data is non-normal, non-parametric models have been employed for this study, which can handle such data without relying on normality assumptions. Table 3Statistical description of the parameters.ParameterMeanStdMinMaxSkewnessKurtosisCement (Kg/m3)366.6890.53132580− 0.70− 0.02Sand (Kg/m3)651.72161.44941047− 0.190.37Coarse aggregate (Kg/m3)1001.26240.431801629− 0.651.64Water (Kg/m3)184.0039.2488370− 0.321.06Natural fiber (Kg/m3)0.520.7703.851.591.95SCM (Kg/m3)5.508.680552.055.72Curing period (days)31.0713.9514903.1010.08Compressive strength (MPa)32.3910.817.757− 0.04− 0.53 Statistical description of the parameters. Contour map of input and output parameters. The Seaborn library was employed to generate the correlation contour maps in Fig. 4, which illustrate the interconnection between the various mix components of the concrete and compressive strength (CS). A high concentration of cement data within the 350–450 kg/m³ range is depicted in Fig. 4a. Figure 4b shows that compressive strength values are most concentrated when fine aggregate content ranges between 650 and 800 kg/m³, indicating this is a commonly adopted range in the mixes. In Fig. 4c, a large group of data points is found between 800 and 1200 kg/m³ of coarse aggregate, where compressive strength tends to peak. Figure 4d illustrates the relation between water content and compressive strength, with the highest data concentration observed between 150 and 220 Kg/m³ of water and 20–50 MPa of strength. The marginal plots indicate that most values cluster around 180–200 Kg/m³ of water and 30–40 MPa of compressive strength. Figure 4e–g illustrates the influence of SCM content, curing age, and natural fiber percentage on compressive strength. Similarly, Fig. 4e reveals two distinct peaks in SCM content, mostly between 5 and 10 kg/m³, each associated with varying strength levels. As shown in Fig. 4f, the compressive strength is primarily concentrated between 20 and 40 days of curing age, indicating the least variation compared to other input parameters, where most of the curing age data clusters around 28 days. Lastly, Fig. 4g indicates that compressive strength is highest when the natural fiber content is near 0%, with a few observations extending beyond 1%. From Fig. 5, the Pearson correlation coefficient between all the pairs of variables was calculated. The coefficient is restricted within the − 1 to 1 range such that the coefficients greater than 0 imply the existence of a positive correlation between two variables, whereas coefficient less than 0 implies a negative correlation between the two variables. The nearer the coefficient is to any of the extremes, the higher the correlation of the features. The heat map shows that 6 of the input variables are positively correlated with Compressive strength (CS), but water and similarly, these data inputs are significantly correlated with water and are less clearly correlated with Compressive strength (CS). Notably, cement shows the highest positive correlation with Compressive strength (R = 0.47), suggesting that the concrete strength shows a direct correlation with this factor. There is a moderate negative relation between the water content and compressive strength (CS), implying that the more water there is, the less strength there is. The main relationship between curing age, fine aggregate, and CS is highly positive, meaning that an increase in the level will increase the strength to the maximum level. Correlation matrix of the relationship between input and output parameters. Error plot and performance evaluation of ML algorithms Figure 6a illustrates the predictive performance of the XGB model in estimating the concrete compressive strength (CS). The figure presents both the experimental and predicted CS values along with the error distribution for the training and testing datasets. During the training phase, the model maintains relatively consistent prediction accuracy, with errors largely contained within 31 MPa. During the testing phase, although the error margin slightly increases, the error range remains approximately 37 MPa, indicating a controlled rise in prediction deviation. Figure 7a provides a scatter plot depicting the relationship between the predicted and experimental compressive strength, where the R² values for the training and testing are 0.9556 and 0.8295, respectively. It means that the training set could be predicted well and that the testing set would be moderately accurate. Figure 6b illustrates the predictive performance of the LGBM model. In the line graph, the predictions from the training data align with the experimental values, and the errors remain within 29 MPa. During the testing phase, the error stays within 27 MPa. The training error range of the data inputs of the LGBM model is slightly lower compared to that of the XGB model training data inputs, while the testing error range is 10 MPa less compared to the XGB model. Strong correlation of predicted and actual values is indicated in Fig. 7b, and the values of R2 are 0.9230 during training and 0.8637 during testing. The LGBM model has a slightly lower prediction performance in comparison with the XGB model during the training phase, while LGB outperforms XGB in R² during the testing phase. It suggests that LGBM demonstrates a more stable and reliable prediction accuracy than the XGB model. Figure 6c compares the observed and predictive values of compressive strength of the NRFC concrete using the RF model. The line plot suggests that the estimated values follow the experimental trends well in the training phase, with most errors remaining within 32 MPa. However, in the testing phase, the error values increase within 30 MPa, suggesting reduced accuracy when exposed to unseen data. The error range is bigger than that of the LGBM model and lower than that of the XGB model during both stages. Relative error of compressive strength for training and testing datasets of (a) XGB, (b) LGBM, (c) RF, (d) KNN, (e) DT, (f) GB models. Relation between predicted vs. observed value for both training and testing stages of the following models: (a) XGB, (b) LGBM, (c) RF, (d) KNN, (e) DT, (f) GB. The scatter plot in Fig. 7c presents the relationship between the predicted and actual values, where training data points are more densely aligned with the expected outcomes than the testing data in the RF model. The R² values for training and testing are 0.9379 and 0.7884, respectively. These results imply that while the RF model learns the training data effectively, its generalization capacity is weaker than that of the LGBM and XGB models, showing higher dispersion and slightly lower testing accuracy. At Fig. 6d, the error plot shows that the predicted values follow the experimental trends for the training and testing datasets employing the KNN model. However, noticeable fluctuations exist throughout, with the error distribution exceeding 50 MPa during the training phase and within 40 MPa during the testing phase, indicating reduced consistency in the model’s generalization. A higher mismatch exists between the observed and predicted results, with both stage errors surpassing those of the XGB, LGBM, and RF models, suggesting relatively weaker predictive reliability. Figure 7d further supports this observation, showing the correlation between predicted and experimental values through a scatter plot using the KNN model. While the R² values of 0.8460 for training and 0.8257 for testing imply moderate model accuracy, especially from the testing set. Though the KNN model shows less accuracy than XGB, LGBM, and RF models during training, its testing accuracy is comparable to XGB, higher than RF, but lower than the LGBM model. As shown in Fig. 6e, the predicted compressive strength values strongly aligned with the experimental data during the training phase using the DT model. Most prediction errors are within 32 MPa, with a significant portion confined to within 5 MPa, indicating a high degree of model accuracy. The error range of the training stage is similar to that of the XGB, LGBM, and RF models. However, in the testing phase, the model demonstrates a clear decline in prediction accuracy, with errors exceeding 55 MPa. This implies that the model has less ability to generalize data and highlights that its error range is worse than that of the XGB, LGBM, RF, and KNN models. Figure 7e further supports this observation through a scatter plot comparing predicted and experimental values in the DT model. The training data points cluster tightly around the ideal diagonal line, yielding a high R² of 0.9499, which implies that the prediction accuracy is superior to the KNN model and comparable to the XGB, LGBM, and RF models during the training stage. On the contrary, the data on testing indicates a large scatter and deviation with a lesser R2 value of 0.7212. These findings indicate that its predictive reliability is greatly diminished when applied to testing data compared with all models. Figure 6f exhibits the error between observed and predicted CS values for training and testing sets in the GB model. The error figure, especially in the testing phase with noticeable fluctuations, reveals that the model sometimes struggles with new data. Errors in the training phase range up to 30 MPa. In contrast, the errors in the testing stage scatter around 45 MPa, which suggests that the GB model is more reliable than the DT model and slightly less reliable than XGB, LGBM, RF, and KNN models. The scatter plot between predicted and experimental values is shown in Fig. 7f using the GB model. In the training stage, most points are close to the diagonal line, like the DT model, which represents perfect prediction. While the training data fits closely, the testing data shows a broader spread, particularly at higher strengths. The R² values of 0.9499 for training and 0.8326 for the testing phase reflect this pattern, that the model learns well during training, but is less accurate on unseen data. The GB model performs reasonably well in the training stage in comparison with XGB, DT, RF, and LGBM models. In the testing stage, this model’s predictive accuracy is less than that of the LGBM model. According to the error plot, the LGBM model shows less error in both the training and testing phases than all other models. Based on the scatter plot, the LGBM model has slightly less predictive accuracy than XGB, RF, DT, and GB during the training period; however, LGBM outperforms all models during the testing stage, indicating superior capability in predicting and generalizing new data. Comparative analysis of ML algorithms The radar graph (Fig. 8), as well as Table 4, shows the evaluation matrix for the machine learning algorithms, including R2, RMSE, MAE, and MAPE values during training and testing, where test performance emphasizes a critical factor in identifying the best model, rather than training accuracy. Figure 8a provides a side-by-side comparison of how six machine learning models performed on both training and testing datasets using R2 values. Among them, LGBM stood out with consistently superior R2 scores of 0.9230 and 0.8637 for both the training and testing datasets. This suggested that the testing stage made accurate predictions and generalized well to new data, while the training stage reduced errors and fit data precisely. On the other hand, the XGB model has an excellent ability to reduce error and fit the dataset, which is indicated by the highest R2 value of 0.956 during the training stage. However, its prediction and generalization ability are much less compared to the LGBM model, with an R2 value of 0.829 in the testing phase. Performance evaluation of all models (a) R2, (b) MAE, (c) MAPE, and (d) RMSE. Table 4Evaluation of the model’s performance indicators.DatasetIndicatorsML techniquesXGBGBRFDTKNNLGBMTrainingR20.9560.9490.9380.9490.8450.923RMSE2.2382.3782.6472.3784.1682.947MAE1.1191.4781.7081.2802.6052.083MAPE0.0390.0520.0600.0450.0860.075TestingR20.8290.8330.7880.7210.8260.864RMSE4.6894.6465.2245.9964.7404.192MAE3.0603.0313.3303.5973.2592.950MAPE0.1150.1140.1270.1290.1160.108 Evaluation of the model’s performance indicators. Other models like GB, RF, and DT performed well during training but dropped sharply on the test set, pointing to overfitting and poor generalization, while KNN showed the weakest performance overall, particularly on the test data, with an R2 value of 0.8257. Figure 8b displays a radar chart comparing the Mean Absolute Error (MAE) of six machine learning models on both datasets. Lower MAE values indicate better performance of the models. From the chart, it’s clear that the LGBM model performed well, showing the lowest MAE on testing data (2.950). In contrast, the DT, RF, GB, XGB, and KNN models had higher MAE values, especially on the testing set, indicating larger average prediction errors and possible overfitting, which points to less accurate predictions compared to LGBM. Figure 8c displays the MAPE values, where lower values indicate better performance. Similarly, the LGBM model performs better, showing the lowest MAPE values of 0.108 among all algorithms, followed by GB (0.114) and XGB (0.115). On the other hand, in Fig. 8d, the RMSE values are displayed, where lower values signify superior performance. Like the MAPE chart, the LGBM model exhibits better performance on testing data, achieving the lowest RMSE value of 4.192, being the top performer, followed by GB (4.646) and XGB (4.689). The results, presented in Table 5, show that all models considered in this study performed within an acceptable range when compared to the models used by other researchers in previous studies. Table 5Model’s performance comparison with previous research models.ModelR2RMSE (MPa)ReferenceXGB0.824.68This studyGB0.834.64RF0.785.22DT0.725.99KNN0.824.74LGBM0.864.19ANN0.894.5649DT0.869.47RF0.866.85SVR0.827.35XGB0.855.97GB0.893.98KNN0.619.25RR0.719.64LR0.718.77ENR0.679.23PCA0.826.87MLR0.921.2577Ridge0.921.26Lasso-0.014.52ENet0.871.57SGD0.811.92RF0.831.83SVR0.950.9GBM0.921.22XGBM0.921.23LGBM0.612.81XGB0.932.8878RF0.903.42GB0.893.56XGB0.944.4379RF0.934.6MLP0.798.43LGBM0.944.26CatBoost0.915.33LGB0.936.1780XGB0.926.91RF0.917.64XGB-RF0.936.47XGB-LGB0.955.25RF0.757.8081GB0.875.70AB0.974.30KNN0.739.40XGB0.865.90KNN0.798.1337XGB0.876.43RF0.837.36 Model’s performance comparison with previous research models. Evaluating performance using the Taylor diagram The Taylor diagram in Fig. 9 visually evaluates the performance of six machine learning models, XGB, LGBM, RF, KNN, DT, and GB, by illustrating their standard deviation and correlation with observed data during training and testing. This tool helps assess how well each model captures the pattern and variability of real-world values. In both phases, all models demonstrate high correlation with the observed data, with R values clustering around 0.95 or higher, indicating strong correlation and effective predictive capability. The break lines (black dashed) indicate the line of constant correlation, which highlights how closely the models match observed patterns. From the training perspective, the two models, LGBM and RF, are aligned closest with the observed reference point (red star), indicating a strong correlation and similar standard deviation. This suggests that these models were able to capture the data’s variability to maintain good predictive performance. In the test stage, the performance trend follows a similar pattern, and LGBM and RF maintain better than other models the original data consistency, as demonstrated by the true time series. Differences are not very noticeable, but LGBM is slightly closer to the reference line, meaning it generalizes slightly better on the test set. Models like XGB and GB also perform well in both phases but show slightly larger deviations from the observed standard deviation. In contrast, KNN and DT are positioned farther from the reference, particularly in terms of standard deviation, suggesting relatively lower precision in capturing the variability of the observed data. The Taylor plot indicates that LGBM and RF are our research’s most stable and consistent models, successfully balancing precision and variability in the training and testing datasets. Taylor plot of all models in this research (a) Training stage, (b) Testing stage. SHapley additive explanations (SHAP) analysis Data on SHAP values have been used to estimate the impact of input features on the model prediction, as shown in Fig. 10. The mean SHAP value denotes the average contribution of each feature in all data samples. Feature importance was evaluated using the LGBM model due to its strong performance in previous phases. Among all input variables, cement content emerged as the most influential factor, with an average SHAP value of 5.73, highlighting its dominant role in enhancing compressive strength (CS). This aligns with its known effect on strength gain through hydration and binding of other mix components. Water content and coarse aggregate followed in importance, with water playing a more critical role due to its influence on hydration, workability, and durability, while coarse aggregate enhances load distribution and reduces shrinkage. SCM ranked next, contributing to strength by improving the concrete’s microstructure and supporting secondary C-S-H formation. Fine aggregate also showed a notable impact by ensuring mix cohesion and density. Finally, fiber content and curing age, though lower in SHAP value, played distinct roles in refining the material’s strength development over time. Feature impact evaluation based on mean SHAP values. SHAP values were used to evaluate the proportional influence of each input variable on the concrete compressive strength prediction; the findings are presented in Fig. 11. The significance of each variable is represented by a cluster of colored dots on the x-axis, where the amount and direction of the influence on the model output are determined by the positive or negative value of SHAP. The y-axis is the ranking of features in terms of the net effect, and a gradient of colors, blue to red, denotes the actual feature value that has been ranked high with red, and in contrast, the low value that has been ranked in blue. From the Fig. 11, cement content is the most impactful feature, consistently contributing positively to strength predictions, especially at higher values, as shown by the cluster of red points with strong positive SHAP values. Water content also plays a significant role, though its influence is mixed. Assessment of input feature impact using the SHAP. In contrast, higher water content may initially aid workability, but it tends to lower strength, as reflected by the spread of positive and negative SHAP values. Coarse aggregate appears similarly impactful, generally enhancing strength but with a nuanced influence depending on its amount. SCM and fine aggregate also show an apparent positive effect on strength, though to a lesser extent, contributing to a denser and more cohesive mix. In contrast, natural fiber and curing period exhibit relatively lower SHAP values, indicating a subtler effect on the model’s prediction result; however, they still contribute positively overall. The distribution pattern and variation in SHAP values across data points emphasize the interactive nature of these features in influencing the model’s prediction of compressive strength in concrete. The heatmap shown in Fig. 12 provides a comprehensive visualization of SHAP values across all features and data instances. This plot captures how each input variable contributes to individual predictions made by the model. The top curve (f(x)) illustrates the model’s output across instances. At the same time, the color gradient represents the extent and trajectory of each feature’s SHAP value, ranging from positive (red) to negative (blue). The cement content shows a strong positive impact on the model’s predictive capacity in the initial portion of the dataset (approximately the first 100 instances), indicated by the intense red coloring, while its contribution becomes more neutral or slightly negative towards the end. In contrast, water content demonstrates a more varied influence, with alternating bands of red and blue, suggesting its dual role in both enhancing and reducing compressive strength depending on the context of other features. The coarse aggregate and SCM features reflect mixed SHAP values throughout the dataset, indicating that their impact on prediction varies widely with different data combinations. On the other hand, fine aggregate, natural fiber, and curing period appear to have a relatively minor influence overall, as shown by the lighter and more neutral color tones, though subtle patterns still emerge that suggest conditional relevance in certain subsets of the data. This heatmap not only highlights the dominant role of cement but also underscores how the contribution of each feature shifts depending on the data instance, offering a detailed view of feature importance in predicting concrete compressive strength. SHAP heat map plot for all datasets. SHAP interaction plots analysis The SHAP interaction plots in Fig. 13 provide valuable insights into how various input variables interact and influence the output. In Fig. 13a, the relationship between cement and water shows a positive effect of cement, reaching around 600 kg/m³, while the water content influences this effect. Building on this, Fig. 13b illustrates that fine aggregate positively affects strength when used between 400 and 600 kg/m³, particularly when paired with higher SCM levels. However, beyond 700 kg/m³, the impact fluctuates, indicating a limited benefit at higher amounts. Figure 13c reveals that as SCM increases from 0 to 50 kg/m³, the SHAP value remains relatively stable. The optimal range appears around 30 kg/m³, where its influence on the model output is consistent. Interestingly, water content subtly interacts with SCM; higher water levels slightly alter its effect on the prediction. Similarly, Fig. 13d highlights a steady increase in SHAP value as Coarse Aggregate rises from 200 to 1600 kg/m³. The most favorable range is between 800 and 1200 kg/m³, where the material strongly influences the model’s output, and further increase of its value reduces the concrete strength. Plots of interaction for input parameters. In contrast, Fig. 13e presents a more variable pattern: the impact of Natural Fiber fluctuates as its content increases from 0 to 3 kg/m³. Nevertheless, its positive influence becomes more pronounced when paired with higher cement levels, with the ideal range falling between 1 and 2 kg/m³. Figure 13f shows a non-linear relationship between the curing period and SHAP value. After an initial increase, the effect stabilizes, with the optimal curing period identified as 5 to 7 days, especially when natural fiber levels are high. Lastly, the interaction between water and fine aggregate is clear in Fig. 13g. Their combined influence is most evident between 150 and 250 kg/m³ of water. In this range, higher acceptable aggregate levels tend to reduce water’s positive effect, while lower levels allow it to contribute more strongly. Beyond this point, the impact of water becomes more consistent and less sensitive to changes in fine aggregate. PDP analysis Figure 14 depicts the influence of each input variable on the compressive strength of concrete. Cement content strongly and predominantly positively influences CS, as shown in Fig. 14a. The CS value steadily increases from approximately 25 MPa to over 42 MPa when cement increases from 200 to 450 kg/m³. This indicates an enhancement of around 68% in strength, emphasizing cement’s critical role in strength. Figure 14b shows a more complex and generally exhibits a negative trend, with the fine aggregate content increasing from 150 to around 1050 kg/m³, the CS drops from about 34.5 MPa to nearly 32.5 MPa, indicating that excessive fine aggregate reduces strength by affecting packing density or increasing water demand. Figure 14c depicts CS increases gradually from around 29 MPa to 36 MPa as the coarse aggregate rises from 200 to approximately 1000 kg/m³, but excessive presence may disrupt the homogeneity and workability of the concrete. According to Fig. 14d, a negative relationship of water content on CS, increasing from around 100 to 200 kg/m³ as CS declines from approximately 35 MPa to 31 MPa. Beyond 200 kg/m3, strength fluctuates, indicating excess water reaches a saturation point in weakening concrete. Figure 14e illustrates that an increase in CS is observed at 0.1 to 0.2 kg/m3 fiber contents, peaking near 34 MPa, then a steady decline in strength occurs, dropping to around 30.5 MPa at 1.5 kg/m³ and remaining flat thereafter. This suggests that excessive fiber content may negatively affect mix uniformity and compaction, reducing strength. In contrast, Fig. 14f shows that SCM clearly influences CS. The strength increases sharply from approximately 35.5 MPa to 36.5 MPa as SCM content rises to around 20 to 30 kg/m³, then stays almost constant. The effect of curing time on compressive strength (CS) is shown in Fig. 14g. CS increases sharply from 28.5 MPa to 32.8 MPa as curing extends from 15 to 28 days, then slowly reaches around 34.3 MPa at 90 days. This trend confirms that while early curing is critical, extended curing still contributes to long-term strength. The recommendations of the results of the PDP analysis are shown in Table 6 below. Input parameter analysis using PDP plots. Table 6Optimized the value of all mixed materials depending on the PDP analysis.MaterialsOptimum valueRemarksCement400–450 kg/m³Strength peaks at 40–42 MPaFine aggregate400–800 kg/m³Excess fines reduce strengthCoarse aggregate800–1000 kg/m³Up to 36 MPa, then fluctuatesWaterw/c ratio ≤ 0.45Higher ratios lower strengthNatural fiber content0.1–0.2 kg/m³Small dose improves; excess weakensSCM20–30 kg/m³Up to 36.5 MPa; stabilizes laterCuring period28–90 daysFast gain at 28 days; gradual after Optimized the value of all mixed materials depending on the PDP analysis. The parametric optimization method (SHAP and PDP) allows optimizing the selection of materials and adjusting their dosage according to scientific arguments rather than trial and error. As an example, the PDP-generated optimum cement, aggregates, natural fibers, and SCM can be used to inform the field engineers to create the optimum mix that will be the strongest, as well as be sustainable. Further, the outcome of the SHAP confirms the relevance of such parameters as the curing period and the water content in the view of materials science by relating them to the hydration kinetics and formation of microstructures. Experimental validation Materials and experimentation To further evaluate the impact of natural fiber on concrete’s mechanical properties, this study carried out an experimental program using coir fiber (CF). The CFs were collected from local coconut processing units located in the Khulna region, Bangladesh. After washing, the fibers were air-dried for 24 h. To enhance the interfacial bond between the hydrophilic coir fibers and the cement matrix, a chemical treatment was performed using a 5% NaOH solution. The fibers were soaked in the solution for 24 h, then rinsed with distilled water to neutralize excess alkali, and finally dried under ambient conditions. The treated fibers were then cut into 25 mm lengths before being added to the concrete mix. Furthermore, a total of 10 experimental mixes were designed with varying percentages of CF (0.25%, 0.50%, 0.75% and 1%) with constant 5% silica fume (SF) as SCM. The particle distribution of the materials used in this study has been illustrated in Fig. 15. Particle distribution curve of mixed materials used in this study. Table 7 represents the mix proportion applied in the present research. The specimens had been cast into cylinders, and the strength had been measured based on compression at 7, 28, and 56 days of the curing process. Furthermore, compressive strength was measured following the standard ASTM C39 testing guideline. In addition, scanning electronic microscopy (SEM) was carried out to qualitatively assess the impact of natural fiber on the concrete matrix. Table 7Mix proportions of SNF concrete.Mix IDMaterials (kg/m3)OPCSFFACAWaterCF (%)S0NF0410066410321780S0NF0.25410066410321780.25S0NF0.5410066410321780.50S0NF0.75410066410321780.75S0NF1410066410321781.00S5NF0389.520.56641032172.50S5NF0.25389.520.56641032172.50.25S5NF0.5389.520.56641032172.50.50S5NF0.75389.520.56641032172.50.75S5NF1389.520.56641032172.51.00OPC = ordinary portland cement Type-I, SF = locally available commercial grade silica fume, FA = fine aggregate having FM of 3.06, specific gravity of 2.74 and Loose unit weight of 1684.4 kg/m3, CA = coarse aggregate having maximum size of 20 mm, specific gravity of 2.94 and Moisture content of 1.82%, CF = treated coir as natural fiber. Mix proportions of SNF concrete. OPC = ordinary portland cement Type-I, SF = locally available commercial grade silica fume, FA = fine aggregate having FM of 3.06, specific gravity of 2.74 and Loose unit weight of 1684.4 kg/m3, CA = coarse aggregate having maximum size of 20 mm, specific gravity of 2.94 and Moisture content of 1.82%, CF = treated coir as natural fiber. Compressive strength of natural coir fiber reinforced concrete Figure 16 illustrates the compressive strength of concrete at 7, 28, and 56 days of curing. The results indicate that the inclusion of different ratios of coir fiber (CF) and a constant 5% of SF resulted in an improvement in compressive strength over time. The mixture containing 5% SF and 0.75% CF had the highest compressive strength at all curing ages. However, increasing the amount of NF reduced the compressive strength and increased the degree of confinement. This reduction could be due to the disruption of the bond structure, increased porosity, and the slowing down of chemical reactions caused by the high amount of CF. Compressive strength of each specimen at the 7, 28, and 56-day curing intervals. The compressive strength of the S5NF0.75 mixture increased by 41.3% after a curing period of 28 days over the control sample. The high strength at this curing age is due to the optimal ratio of SF and CF, which enhances the microstructure and bonding of the material. However, excessive amounts of NF can adversely affect the performance of the concrete. Excess fiber content can increase voids, reduce the effect of chemical bonding, and result in a weaker concrete. Such behavior indicates the need to optimize the amount of natural fiber so that the benefits of both fiber and SCM are realized during concrete mix design. Microstructural assessment The microstructural evaluation of NFRC was carried out to better understand the role of natural fibers in modifying the inner matrix of the cementitious composite. Figure 17a and b show natural fibers under secondary electrons (SE) and backscattered electrons (BSE) imaging modes, respectively. The fibers exhibit elongated rough external surfaces, which are essential for mechanical interlocking within the cementitious matrix. High-magnification views in Fig. 17c and d reveal surface microgrooves and irregularities. These irregular surface characteristics, ensuring better adhesion and limiting fiber pull-out under stress conditions, contribute to the overall mechanical performance by enhancing bond strength and reducing the risk of failure underloading. Furthermore, SEM images of the control mix matrix in Fig. 17e and f show a relatively dense matrix with a visible interfacial transition zone (ITZ), where aggregate and cement paste are well integrated. In contrast, Fig. 17g and h show the fiber-concrete composite samples. The presence of bridging fibers spanning across microcracks demonstrates the fibers’ ability to arrest crack propagation and delay failure. Additionally, fiber pull-out marks and gaps around embedded fibers indicate partial debonding, which contributes to energy dissipation under loading. These microstructural features are consistent with the predictive modeling results, where fibers were found to be significant predictors of compressive strength, especially in terms of improving crack resistance and limiting stress propagation. The microcracks appear finer and more tortuous in the fiber-reinforced concrete mixes, which suggests that the inclusion of natural fibers contributes to enhanced crack resistance and improved post-cracking behavior. This observation correlates with the machine learning model’s prediction, where fiber content was found to improve strength up to an optimal point, aligning with experimental findings. Studies conducted by Sobuz, Aayaz82 using hybrid fiber combinations in concrete composites reported similar findings, where the incorporation of fibers within the matrix significantly enhanced crack-bridging capacity and improved the post-cracking toughness of the material, which is well captured by the predictive model in these studies. SEM micrographs of (a) Fiber in SE mode, (b) Fiber in BSE mode, (c) Fiber surface in SE mode, (d) Fiber surface in BSE mode, (e) Control matrix in SE mode, (f) Control matrix in BSE mode, (g) Specimens having NF in SE mode and (h) Specimens having NF in BSE mode. Conclusions Machine learning is transforming civil engineering with the potential to optimize and predict concrete properties and data-driven innovations of construction processes. Machine learning models were applied to the research in the prediction of the CS of concrete. This study aimed to contrast the predictive accuracy and effectiveness of six models that predict CS using basic input parameters. The following conclusions are drawn based on the results of our study on the prediction of compressive strength of NFRC mixes: The regression analysis indicates that the LGBM model performs best and is the most reliable model among the other five machine learning models. Of them, LGBM obtained the largest R² score of 0.864 in the testing period and R2 score of 0.923 in the training stage. This suggests solid predictive accuracy and generalization capability in estimating strength and low overfitting of the dataset. The XGB, GB, RF, and DT outperformed in the training phase, with the R2 scores of 0.956, 0.949, 0.938, and 0.949, respectively, indicating low overfitting ability. However, their R2 scores decreased to 0.829, 0.833, 0.788, and 0.721 in the testing phase, indicating lower predictive accuracy and generalization capability. The regression analysis indicates that the LGBM model performs best and is the most reliable model among the other five machine learning models. Of them, LGBM obtained the largest R² score of 0.864 in the testing period and R2 score of 0.923 in the training stage. This suggests solid predictive accuracy and generalization capability in estimating strength and low overfitting of the dataset. The XGB, GB, RF, and DT outperformed in the training phase, with the R2 scores of 0.956, 0.949, 0.938, and 0.949, respectively, indicating low overfitting ability. However, their R2 scores decreased to 0.829, 0.833, 0.788, and 0.721 in the testing phase, indicating lower predictive accuracy and generalization capability. Based on the performance analysis, the LGBM model is the most reliable alternative in estimating the compressive strength of NFRC since it is the most effective among all other models in all aspects of lowest error measures (test RMSE: 4.192, MAE: 2.950, MAPE: 0.108%), and maximum predictive accuracy. Based on the performance analysis, the LGBM model is the most reliable alternative in estimating the compressive strength of NFRC since it is the most effective among all other models in all aspects of lowest error measures (test RMSE: 4.192, MAE: 2.950, MAPE: 0.108%), and maximum predictive accuracy. According to the Taylor diagram analysis, all six models are working well, with the correlations taking a similar value of around 0.95 with similar standard deviations between training and testing. Among the above, LGBM and RF have a minor performance advantage because they are closer to the reference point, which suggests greater stability and generalization. The LGBM in this case is the most consistent with the observed data, so it is the best one to predict the NFRC compressive strength. According to the Taylor diagram analysis, all six models are working well, with the correlations taking a similar value of around 0.95 with similar standard deviations between training and testing. Among the above, LGBM and RF have a minor performance advantage because they are closer to the reference point, which suggests greater stability and generalization. The LGBM in this case is the most consistent with the observed data, so it is the best one to predict the NFRC compressive strength. SHAP analysis demonstrated that the features that had the most significant influence on the prediction of the CS of the concrete were cement, water, and coarse aggregate. Their significant positive contribution is indicated by their respective mean SHAP values of + 5.73, + 2.05, and + 1.91, due to their management of hydration, porosity, and transfer of loads in controlling concrete compressive strength. SHAP analysis demonstrated that the features that had the most significant influence on the prediction of the CS of the concrete were cement, water, and coarse aggregate. Their significant positive contribution is indicated by their respective mean SHAP values of + 5.73, + 2.05, and + 1.91, due to their management of hydration, porosity, and transfer of loads in controlling concrete compressive strength. The analysis of the PDP indicates that the strength of NFRC is the highest at the cement ratios of 400–450 kg/m3, SCM ratios of 20–30 kg/m3, and the curing period of no more than 90 days, with the assistance of the balanced ratios of coarse and fine aggregates. However, excess water (w/c > 0.45) and dosage of fiber above 0.2 kg/m3 decrease strength, indicating the proportionality between these two factors to attain durable and high-performance mixes. The analysis of the PDP indicates that the strength of NFRC is the highest at the cement ratios of 400–450 kg/m3, SCM ratios of 20–30 kg/m3, and the curing period of no more than 90 days, with the assistance of the balanced ratios of coarse and fine aggregates. However, excess water (w/c > 0.45) and dosage of fiber above 0.2 kg/m3 decrease strength, indicating the proportionality between these two factors to attain durable and high-performance mixes. Experimental assessment further confirmed that the incorporation of natural fiber up to 0.75% gradually increased the mechanical strength of concrete, while further addition led to a decline in performance. Experimental assessment further confirmed that the incorporation of natural fiber up to 0.75% gradually increased the mechanical strength of concrete, while further addition led to a decline in performance. The outcomes were also verified by the SEM micrograph, where the fiber helps to significantly restrain microcrack propagation and enhance fiber-bridging mechanisms within the cementitious matrix. The outcomes were also verified by the SEM micrograph, where the fiber helps to significantly restrain microcrack propagation and enhance fiber-bridging mechanisms within the cementitious matrix. The proposed technique provides a reliable and efficient method of predicting the compressive strength of NFRC using machine learning models. It can greatly limit the amount of time and expense involved in traditional lab testing. Moreover, the knowledge learned from this study may help in the mix design optimization and promote sustainable large-scale production of fiber-reinforced concrete in the construction industry. Limitations and further studies Certain limitations of this study can be addressed in future investigations. Due to the availability of materials and laboratory facilities, time and cost limitations, the experimental validation was done only for coir fiber, which limits the validation for jute and bamboo fibers. In this sense, there is yet a lack of generalizability of the results for all types of NFRCs. The data set was collated from a variety of literature sources, which introduced heterogeneity in mix proportions and testing conditions that may impact predictive performance. In addition, the scope was limited to compressive strength, without considering other mechanical and durability properties, and no field-scale validation was carried out. To address this issue, future investigation can be done by including a hybrid fiber system, durability and other mechanical properties, field-scale validation, and hybrid supervised AI techniques. This will make it possible to find out a more reliable predictive model for the infrastructure industry. Acknowledgements The authors extend their appreciation to the Deanship of Research and Graduate Studies at King Khalid University for funding this work through large group project under grant number (RGP2/33/46). Author contributions P.P.: Conceptualization, Methodology, Formal analysis, Data curation, Investigation, Visualization, Writing—original draft, Writing— review & editing. M.K.I.K.: Conceptualization, Methodology, Formal analysis, Investigation, Supervision, Resources, Writing—original draft, Writing—review & editing. R.D.G.: Investigation, Validation, Methodology, Data curation, Writing—original draft, Writing—review & editing. Md. H.R.S.: Methodology, Formal analysis, Data curation, Writing—original draft, Writing—review & editing. P.B.: Formal analysis, Data curation, Writing—original draft, Writing—review & editing. M.J.: Methodology, Writing—original draft, Writing—review & editing. S.A.A.: Data curation, Writing—original draft, Writing—review & editing. Data availability The dataset, as well as the reference sources of the collected dataset, can be found in the open repository – 10.17632/6jxy89krrx.1. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Md. Kawsarul Islam Kabbo, Email: kabbo@becm.kuet.ac.bd. Sani Aliyu Abubakar, Email: saliyu@kiu.ac.ug. References Associated Data Data Availability Statement The dataset, as well as the reference sources of the collected dataset, can be found in the open repository – 10.17632/6jxy89krrx.1.