Mechanical and flexural fatigue behaviour of steel fibre reinforced RAP-based concrete: experimental and machine learning investigation Authors: Yeswanth Paluri, Sri Ram Ravi Teja Prathipati, Siva Shanmukha Anjaneya Babu Padavala, Adamu Mulatu Kumara, Heeralal Mudavath, Rathish Kumar Pancharathi Published: 2026 ABSTRACT AbstractThis study examines the mechanical and flexural fatigue behaviour of concrete with reclaimed asphalt pavement (RAP) aggregates with and without steel fibre reinforcement. The use of RAP as partial replacement for natural aggregates represents a sustainable solution for reduction of construction waste and saving of natural resources. Twenty concrete mixes were prepared with RAP replacement levels varying between 0 and 30% and steel fibre contents varying between 0 and 1.25% by volume. The experimental investigation comprised t evaluation of strength characteristics, stress–strain behaviour, and flexural fatigue performance. Results show that increase in RAP content leads to a gradual decrease in the strength of the concrete due to weaker interfacial transition zone formed between the RAP aggregates and the cementitious matrix. However, the addition of steel fibres significantly enhances the tensile and flexural properties; the optimum steel fibre dosage was found to be 1%. At this dosage the compressive toughness was significantly increased, and the split tensile and flexural strengths improved as much as 45% and 48%, respectively. Stress–strain behaviour of RAP-based fibre reinforced concrete under compression is described by a modified Saenz equation based stress–strain model. The results of flexural fatigue tests indicated that fibre reinforcement substantially improves the fatigue life and endurance limit of RAP-based concrete. Machine learning models based on random forest regression and gradient boosting regression were also developed for the prediction of mechanical properties with the highest predictive accuracy of the gradient boosting model (R2 = 0.96). The integrated experimental and machine learning framework developed in this study provides a preliminary proof-of-concept methodology for evaluation and optimization of sustainable RAP-based fibre reinforced concrete for pavement applications.Keywords: Reclaimed asphalt pavement (RAP), Steel fibre reinforced concrete (SFRC), Machine learning, Flexural fatigue, Endurance limit, Sustainable concreteSubject terms: Engineering, Materials science Introduction Concrete is the most widely used material in the world second only to water because it can be used in many different applications for infrastructure, due to its versatility, durability and adaptability. As a composite material, the properties of concrete are greatly affected by the quality of the materials which make it up. Concrete generally contains cementitious materials, water, coarse aggregates and fine aggregates. The environmental impact of production of these materials is significant, and it is, therefore, wise to investigate sustainable alternatives. The use of waste/recyclable material as an alternative for the conventional materials is being widely studied as it not only decreases the amount of waste being dumped in the landfills but also decreases the impact to the environment due to extraction and handling of the natural materials1,2. When used effectively, recycled materials have the potential to be better and cost-effective construction resources3. Given that aggregates make up more than 70% of the total volume of the concrete, the incorporation of alternative or waste-derived materials as aggregates can make a huge difference in the sustainable practice of construction. Extensive research has established that a range of waste materials, such as construction and demolition waste (C&DW), reclaimed asphalt pavement (RAP) and discarded rubber tires can be effectively utilized as a partial replacement of coarse aggregates in concrete4–12. The use of these waste materials encourages sustainable building and the principles of circular economy, while at the same time solving waste management problems. In particular, reclaimed asphalt pavement (RAP) has attracted increasing interest as a possible aggregate substitute in concrete applications. RAP is produced in the rehabilitation and reconstruction work of pavement facilities, where existing asphalt pavements are milled or removed in order to restore structural performance or enhance road serviceability13. The resulting material usually takes the form of aggregates covered with aged asphalt binder. RAP has been widely used in asphalt mixtures because of its benefits in economics and environment. However, a significant amount of RAP is still being underutilised and this leaves substantial quantities of RAP remain stockpiled or are disposed of in landfills14. The combination of these issues, which include the increasing scarcity of natural aggregates, and the rising costs of landfill disposal creates a need to look for alternative uses of RAP materials. Once properly processed (crushing, washing and screening), RAP can be used as a partial substitute for natural aggregates in concrete. RAP aggregates generally comprise of mineral aggregates that are coated with aged asphalt binder and fine dust particles. Nevertheless, the aged asphalt binder which is present on the RAP particles poses problems when used in cement-based material. The asphalt coating creates a rather weak and hydrophobic interface with the surrounding cementitious matrix which may be attributed to weaker interfacial transition zone characteristics and therefore diminished mechanical strength. Unlike natural aggregates, RAP particles are coated with aged asphalt binder, which creates a hydrophobic surface and reduces chemical bonding with the cement paste. Consequently, RAP concrete generally develops a weaker interfacial transition zone compared to conventional concrete containing natural aggregates. Early studies on the use of RAP in Pavement Quality Concrete (PQC) were performed by Delwar et al.15, who noted that the hardened properties of RAP-based concrete were inferior and as such, its application was limited to non-structural applications. Similar results were later reported by Hassan et al. and Al-Oraimi et al.16,17. Further studies have repeatedly shown that although RAP aggregate reduces the strength of the concrete, it can be used as a partial substitute for NA18–26. The degradation of the mechanical performance of RAP-based concrete, compared to conventional concrete, is mainly attributed to weaker interfacial transition zone characteristics between the cementitious matrix and the asphalt-coated RAP aggregates. In order to reduce this problem, several techniques have been suggested such as the pre-treatment of RAP by abrasion and attrition treatment processes or the use of mineral admixtures to improve mechanical performance27–33. For the design of concrete pavements and other structural applications exposed to repeated traffic loads, mechanical strength is not enough to guarantee long-term durability. Instead, the fatigue performance of the material becomes an important design parameter. Fatigue failure occurs when concrete undergoes cycles of loading below its ultimate strength and forms progressively widening micro-cracks within the concrete34. Over time, such micro-cracks accumulate and eventually cause structural failure. Therefore, it is imperative to improve the fatigue resistance of concrete to improve the durability and service life of pavement structures, such as highways, airport runways, bridge decks, etc. Fatigue strength, also known as the endurance limit, is the maximum stress level that a material can withstand for a given number of loading cycles without breakdown, usually taken as being about two million cycles for concrete pavements35. One useful way of enhancing fatigue strength of concrete is to use fibres. Fibre reinforcement improves the mechanical behaviour of concrete by bridging the crack, retarding the propagation of the crack, and allowing better energy absorption capacity. Steel fibre reinforced concrete (SFRC) in particular has been well studied and widely used in pavement and structural applications because of its excellent crack control and toughness characteristics36–42. The random distribution of steel fibres in the concrete matrix gives them an extra resistance to crack growth by mechanisms such as fibre pull-out and fibre bridging, leading to an improvement in the tensile and flexural performances43. Several experimental studies have shown tremendous improvements in the fatigue life and endurance limit of concrete with the addition of steel fibres44–48. While the fatigue behaviour of recycled aggregate concrete (RAC) from construction and demolition waste has been extensively studied, investigations on the flexural fatigue behaviour of RAP-based concrete, especially in combination with steel fibre reinforcement, are not abundant. Previous studies have demonstrated that if more recycled aggregate is used, fatigue life could be reduced due to the weaker interfacial bonding as well as the increased material heterogeneity49–52. However, the impact of RAP aggregate on fatigue performance, especially in conjunction with steel fibre reinforcement, has not been widely investigated. Given the growing importance being attached to the use of sustainable pavement materials, the need to understand the fatigue behaviour of RAP-based concrete is critical in determining its potential application to pavement applications. In addition to experimental investigations, recent development of machine learning (ML) has made it possible to develop predictive models that can capture complex relationships between material composition and concrete performance. Machine learning approaches have been widely used in civil engineering for predicting mechanical properties of concrete mixtures, optimizing the mix design parameters, and decreasing the experimental effort. The combination of machine learning methods and experimental investigations offers an efficient framework for modelling the behaviour of sustainable concrete materials and data-driven decision-making for mix design optimization53–55. However, the reliability of machine learning models strongly depends on the quality and size of the experimental dataset. For relatively small experimental datasets, the developed models should be interpreted as preliminary predictive tools intended to capture experimental trends within the investigated parameter range. Although previous studies have investigated the mechanical or fatigue behaviour of RAP-based concrete and the influence of steel fibre reinforcement independently, limited studies have integrated these aspects within a unified framework. Furthermore, studies focusing on the combined evaluation of flexural fatigue behaviour, stress–strain modelling, and machine learning-based prediction of steel fibre reinforced RAP concrete remain scarce. In particular, the earlier work reported by Paluri et al.22 primarily focused on the flexural fatigue behaviour of steel fibre reinforced RAP concrete. The present study significantly extends that work by incorporating comprehensive stress–strain modelling, compressive behaviour characterization, analytical constitutive modelling, and machine learning-based prediction frameworks within an integrated experimental methodology. Therefore, the novelty of the present study lies in integrating experimental investigation, analytical stress–strain modelling, fatigue assessment, and machine learning techniques within a single framework for evaluating RAP-based fibre reinforced concrete for pavement applications. Therefore, the present study is aimed to investigate the mechanical and flexural fatigue behaviour of concrete with reclaimed asphalt pavement aggregates with and without steel fibre reinforcement. In the scope of the research activities an evaluation of the influence of RAP replacement levels and fibre dosage on compressive strength, split tensile strength, flexural strength, stress–strain behaviour and fatigue performance is performed. The present investigation is limited to a single concrete grade, one RAP source, selected RAP replacement levels, and one type of steel fibre reinforcement. Therefore, the findings should be interpreted within the investigated experimental conditions. Furthermore, analytical stress strain models are developed to describe the compressive behaviour of steel fibre reinforced RAP-based concrete (SFRRAPC). In addition, machine learning methods, namely random forest regression and gradient boosting regression, are used to develop the predictive models of mechanical properties of RAP-based concrete as a function of the mix design parameters. The integrated experimental, analytical and machine learning framework presented in this study are aimed at providing a comprehensive methodology for evaluating and optimizing sustainable RAP-based fibre reinforced concrete for pavement applications. Research significance As infrastructure development and pavement rehabilitation projects have grown in number and size, the quantity of natural aggregates used and reclaimed asphalt pavement (RAP) materials produced has gone up. If RAP can be used in concrete, it can help to decrease the amount of waste sent to landfill sites and conserve natural resources. The use of RAP in practice is still limited, however, due to the poorer interfacial transition zone (ITZ) between the asphalt-coated RAP particles and the cementitious matrix, affecting mechanical performance. This study addresses this challenge by examining how the inclusion of RAP and steel fibre reinforcement affect the behaviour of concrete for pavement use. Steel fibres enhance the performance of concrete by providing crack-bridging action, increased energy absorption and delay in crack propagation, which offsets the performance decrease due to the RAP content. This study measures the stress–strain characteristics and flexural fatigue behaviour, both of which are important for concrete under repeated traffic loading, in addition to evaluating the conventional mechanical properties. In addition, the present study combines experimental research with analytical modelling and machine learning methods to create a comprehensive approach for assessing RAP concrete. The results of this research work will help to enhance the knowledge about sustainable RAP-based fibre reinforced concrete and provide valuable inputs for the design of sustainable and environment friendly fibre reinforced concrete. Experimental study The experimental program consisted of preparation, casting and evaluation of some concrete specimens, in which NA were partially replaced by RAP aggregates. The present study investigated the effect of RAP incorporation systematically on the mechanical properties, stress–strain behaviour and also on the flexural fatigue behaviour of concrete to provide crucial information on its feasibility for pavement applications. Materials The materials used in this study were chosen to comply with the relevant specification of the Indian Standard and also based on consistent performance of the concrete. Ordinary Portland Cement (OPC), 53 Grade was used as the main cement giving a specific gravity of 3.12 and as per IS 12269-201356. Well graded river sand as per the grading requirements of Zone II, with specific gravity of 2.68 and satisfying the requirements of IS 383-201657, was used as the fine aggregate, thereby ensuring adequate workability and strength development. Crushed granite with specific gravity equal to 2.78 and maximum size of 20 mm was used as natural coarse aggregate (NA) considering IS 383-201657. To improve the workability, high range water reducing admixture (HRWRA), Complast 430 based on Sulphonated Naphthalene Formaldehyde (SNF) condensate having specific gravity of 1.22 was used in the mix. Crimped steel fibres with an aspect ratio of 50 and a diameter of 0.5 mm, 1145 MPa tensile strength and 205 GPa modulus of elasticity were added to improve the toughness and flexural performance. The RAP aggregates used in this study were derived from a local asphalt pavement rehabilitation project in which the existing pavement layers were milled during pavement rehabilitation works. The collected RAP material was processed through crushing, washing and air drying before it was used in concrete. RAP particles were washed with a high-pressure water jet to remove loose fines, dust particles and contaminants. This treatment enhanced the bond between the cement paste and the surface of the aggregate and hence enhancing the mechanical performance of RAP-based concrete. Figure 1 shows the RAP aggregates before and after the washing process. To ensure experimental consistency, RAP aggregates ranging from 4.75 to 20 mm were selected and incorporated as a partial replacement for NA in the concrete mix. The gradation curves of RAP aggregates and NA are as shown in Fig. 2. (a) Old pavement being milled; (b) RAP stockpile; (c) RAP aggregates before and after washing. Gradation curves of NA and RAP aggregates. The adhered asphalt content of RAP aggregates was determined using the centrifuge extraction method in accordance with ASTM D2172-1758. In this procedure, the asphalt binder is dissolved using an organic solvent and separated through centrifugation to quantify the residual binder content. The asphalt content of the RAP aggregates used in this study was found to be approximately 4% by weight. Mix proportions and designation A 40 MPa concrete mix with RAP aggregates as a partial replacement for natural coarse aggregates was designed in accordance with IS: 10262-201959. The optimised mix ratio was 1:2:2.5 with a water-cement ratio of 0.36. A high range water reducing admixture (HRWRA), Complast 430, was used at 1% by weight of cement for all the mixes. RAP replacement varied from 0 to 30%, while steel fibre dosage ranged from 0 to 1.25%. The steel fibres in this study were added as a percentage by volume of concrete. The fibre dosages of 0.5%, 0.75%, 1.0%, and 1.25% refer to the volume fraction (Vf) of steel fibres with respect to the total volume of concrete mix. The study assessed the mechanical performance and fatigue behaviour of RAP-based fibre reinforced concrete, with Table 1 detailing the mix compositions. Table 1Mix designation.RAP content (%) ↓Fibre dosage (%) →00.50.751.001.250NAC/RAP0RAP0-SF0.5RAP0-SF0.75RAP0-SF1RAP0-SF1.2510RAP10RAP10-SF0.5RAP10-SF0.75RAP10-SF1RAP10-SF1.2520RAP20RAP20-SF0.5RAP20-SF0.75RAP20-SF1RAP20-SF1.2530RAP30RAP30-SF0.5RAP30-SF0.75RAP30-SF1RAP30-SF1.25 Mix designation. Testing procedure The physical properties of RAP aggregates and natural aggregates were determined in accordance with IS: 2386-201660. After casting, all specimens were demoulded after 24 h and cured in potable water at a temperature of 27 ± 2 °C until the testing age. The workability of fresh concrete was evaluated using the slump test, which evaluates the consistency and workability of fresh concrete, following IS: 1199-201861. Slump values were recorded immediately after mixing to assess the influence of RAP aggregates and steel fibre dosage on fresh concrete workability. The hardened properties were determined per IS: 516-202162 using a 1000 kN servo-controlled testing system. For strength assessment, cube specimens of size 100 mm × 100 mm × 100 mm were tested for compressive strength, while cylindrical specimens of 100 mm diameter × 200 mm height were cast to evaluate split tensile strength. Beam specimens of size 100 mm × 100 mm × 500 mm were used for flexural strength determination. For each mix proportion, three replicate specimens were tested for each mechanical property, and the average value was reported as the representative strength. The coefficient of variation (COV) remained below 10%, indicating acceptable repeatability and consistency of the experimental results. The relatively low scatter observed in the experimental data indicates stable behaviour of the tested mixtures within the investigated RAP replacement levels and fibre dosages. To visually represent variability, error bars corresponding to the standard deviation from the mean have been included. Additionally, cylindrical specimens of size 100 mm diameter × 200 mm height were prepared to evaluate the stress–strain behaviour of concrete, and stress–strain curves were obtained using the average response of three identical specimens. The flexural fatigue behaviour of RAP-based concrete was investigated using beam specimens of size 100 mm × 100 mm × 500 mm under four-point loading to ensure a constant bending moment region and subjected to cyclic loading. The fatigue tests were performed on the 1000 kN servo-controlled dynamic testing system shown in Fig. 3. Cyclic loading was applied at three stress levels corresponding to 0.9, 0.8, and 0.7 of the ultimate flexural strength of the concrete beams. These stress levels were selected to represent the commonly adopted stress range used in fatigue evaluation of pavement quality concrete subjected to repeated traffic loading. A fixed-amplitude sinusoidal loading waveform was applied at a frequency of 2 Hz, representing loading frequencies typical of vehicular traffic on concrete pavements49,63,64. The number of load cycles required to cause failure of the specimen was recorded as the fatigue life of the concrete beam. The experimental setup used for fatigue testing is illustrated in Fig. 3. 1000 kN servo-controlled dynamic testing machine test setup. Results and discussion Physical properties of aggregates Table 2 indicates that RAP aggregates possess slightly lower specific gravity and bulk density compared to NA. The lower specific gravity of RAP aggregates can be attributed to the presence of adhered asphalt binder coating the aggregate surface, which possesses a lower density compared to mineral aggregates. RAP aggregates showed lower water absorption, the lower water absorption of RAP aggregates is attributed to the hydrophobic asphalt binder coating, which limits moisture penetration. In terms of particle shape, RAP aggregates exhibit fewer elongated and flaky particles than NA, with a combined flakiness and elongation index of 19%, well within the 40% limit specified by IS 383:201657. The aggregate impact value and crushing value of RAP aggregates were slightly less than that of NA; this can be primarily attributed to the presence of a stiff, aged bitumen coating on the RAP particles. This stiff bitumen layer acts as a protective shell, providing partial confinement to the aggregate particles and effectively distributing the applied loads during impact and crushing tests. Although RAP aggregates demonstrate higher impact and crushing resistance, they exhibit lower abrasion resistance than NA. The elevated Los Angeles abrasion value in RAP aggregates is primarily due to asphalt-coated agglomerated particles, which are more susceptible to surface wear and degradation under mechanical action. Similar results were observed by other researchers65–67 Table 2Physical Properties of natural aggregates and RAP aggregates.PropertiesNatural aggregatesRAP aggregatesBulk density (kg/m3)1.491.37Void ratio0.790.90Specific gravity2.682.6Water absorption1.201.04Flakiness index2210Elongation index189Angularity number10.2010.78Aggregate impact value18.2014.50Aggregate crushing value18.0115.76Los Angeles abrasion Value24.1038.20 Physical Properties of natural aggregates and RAP aggregates. Fresh properties of RAP-based concrete Figure 4 illustrates the slump values of concrete mixes with varying RAP contents and fibre dosages, demonstrating a progressive decline in workability as RAP content increases. This trend is consistent across both plain and fibre reinforced specimens. Slump of RAP-based concretes with and without steel fibres. The reduction in workability is primarily due to the high viscosity of the asphalt binder coating the RAP aggregates, which restricts particle mobility within the concrete matrix, leading to decreased flowability and compaction efficiency. Additionally, the incorporation of steel fibres further reduces workability, irrespective of the RAP content. This behaviour is attributed to interlocking effect of fibres and fibre clustering which increases internal friction and restricts the flowability of the mix. Despite the observed reduction in workability, the slump values remain within acceptable limits for most practical concreting applications, ensuring adequate placement and compaction. However, HRWRA can be used to improve the workability where required. Mechanical properties of RAP-based concrete Mechanical properties of RAP-based concrete were found for various replacements of natural coarse aggregates with RAP aggregates for various fibre dosages. The variation of different mechanical properties of steel fibre reinforced RAP-based concrete (SFRRAPC) is as discussed below: Compressive strength Figure 5 illustrates the compressive strength variation in RAP-based concrete mixes, showing a gradual decline in strength with increasing RAP content. The reduction in strength may be attributed to the relatively weaker interfacial characteristics between the asphalt-coated RAP aggregates and the surrounding cementitious matrix, as reported in previous studies. The aged asphalt binder coating on RAP particles forms a non-reactive and hydrophobic layer, which impedes proper bonding with the surrounding cementitious matrix. This weak interface is a preferential route for the initiation and propagation of micro-cracks, particularly under tensile and cyclic loading, thus accelerating mechanical degradation. Similar compressive strength reductions because of poor interfacial bonding between RAP aggregates and cement paste have been reported by other researchers25,26,68,69. However, concrete of 20% RAP fulfilled the compressive strength requirement of 40 MPa of pavement quality concrete (PQC) as per MoRTH specifications. The addition of steel fibres increased the compressive strength, with improvement being observed up to 1% fibre dosage beyond which a decrease was observed. The reduction in performance beyond 1% steel fibre dosage is primarily attributed to fibre agglomeration and balling effects, which reduce workability, hinder uniform fibre dispersion, and create localized weak zones within the concrete matrix. The enhancement in compressive strength for SFRRAPC was approximately 1.5–3%, 3.5–6%, 8–12%, and 5–7% at steel fibre dosages of 0.5%, 0.75%, 1.0%, and 1.25%, respectively. A steel fibre dosage of 1% was found to be optimal for balancing strength enhancement, workability, and homogeneous fibre dispersion. The variation observed in the compressive strength values remained within acceptable experimental limits, and no anomalous trend reversal was observed across the investigated RAP replacement levels and fibre dosages. Compressive strength variation of RAP-based concrete with and without steel fibres. Flexural strength and split tensile strength Figures 6 and 7 show the variation of flexural and split tensile strength for the various RAP-based concrete mixes, where the trends are similar to that of compressive strength, that is, with increase in RAP content, the strength is gradually reduced. However, concrete with 30% RAP achieved the pavement quality concrete (PQC) specification of 4.5 MPa flexural strength according to MoRTH specifications. The addition of steel fibres resulted in the substantial improvement of both flexural and split tensile strength with the greatest improvement observed at fibre dosages up to 1%, beyond which only marginal gains were achieved due to the agglomeration of fibres and limited dispersion efficiency. Beyond this limit, further fibre addition resulted in a reduction in strength by virtue of fibre balling, which interferes with uniformity of dispersion within the concrete matrix. The split tensile strength improvement was found to be 17–19% for steel fibre reinforced RAP-based concrete (SFRRAPC) of fibre dosage of 0.5%, 27–31% for 0.75%, 42–50% for 1.0% and 33–40% for 1.25%. Similarly, the flexural strength improvements were recorded to be 9–14%, 24–26%, 41–48% and 29–38% for 0.5%, 0.75%, 1.0% and 1.25% fibre dosages respectively. Similar observations have been made by previous researchers as well25,70–74. These results indicate that 1% steel fibre content is the optimal dosage for maximizing the flexural and tensile performance of RAP-based concrete while maintaining material integrity. The reduction in performance beyond 1% steel fibre dosage is primarily attributed to fibre agglomeration and balling effects, which reduce workability, hinder uniform fibre dispersion, and create localized weak zones within the concrete matrix. The scatter observed in the flexural strength and split tensile strength values remained within acceptable experimental limits, and no anomalous trend reversal was observed across the investigated RAP replacement levels and fibre dosages. Flexural strength variation of RAP-based concretes with and without steel fibres. Split tensile strength variation of RAP-based concretes with and without steel fibres. Machine learning prediction of mechanical properties In addition to the experimental investigation, the machine learning (ML) techniques were used to develop the predictive models for estimating the mechanical properties of RAP-based FRC. Machine learning approaches are increasingly being adopted in materials engineering because of their ability to capture complex nonlinear relationships between mixture parameters and material performance. In the present study, ML models were developed for predicting the compressive strength, split tensile strength and flexural strength of RAP-based concrete using experimental data obtained from the experimental program. Dataset and input parameters The machine learning dataset was derived from the experimental results based on the twenty concrete mixtures with varying RAP replacement levels and steel fibre contents. The mixtures contained RAP replacement levels between 0 and 30% and dosages of steel fibre between 0 and 1.25% by volume. The experimental results of compressive strength, split tensile strength, and flexural strength were considered as the output variables for the ML models. The dataset was split into training and testing data sets where 70% of the data was used for training the model, while 30% of the data was used for testing the model. The predictive power of the models was tested with the aid of typical statistical performance metrics such as mean squared error (MSE), root mean squared error (RMSE), and the coefficient of determination (R2). It should be noted that the present dataset consists of 20 experimental mix combinations, which represents a relatively limited dataset size for machine learning applications. Therefore, the developed machine learning models are intended primarily for exploratory prediction within the investigated RAP replacement and fibre dosage ranges rather than for universal generalization. Considering the relatively limited dataset size, the present machine learning investigation should be interpreted as a preliminary proof-of-concept study intended to evaluate the feasibility of ensemble learning techniques for RAP-based concrete prediction. Although the developed models demonstrated satisfactory agreement with experimental observations, broader datasets covering wider material combinations would be required for establishing generalized predictive capability. Future studies may incorporate leave-one-out cross-validation (LOOCV) or k-fold cross-validation approaches to further improve statistical robustness. Random forest regression (RFR) Random forest regression (RFR) is a type of ensemble learning technique that creates multiple decision trees and integrates their predictions to enhance the accuracy of the model and mitigate the overfitting issue. In the present study, the RFR model was trained on the experimental dataset for the prediction of mechanical properties of RAP-based concrete mixtures. The performance of the RFR model was evaluated on the testing data set. For compressive strength prediction, the model showed mean squared error (MSE) of 1.688, root mean squared error (RMSE) of 1.299, and a coefficient of determination (R2) of 0.934 showing a high correlation between the predicted and experimental values. Similarly, the prediction of split tensile strength gave an MSE value of 0.078, RMSE value of 0.280, and the R2 value of 0.866 showing satisfactory predictive performance. For flexural strength, the model gave an MSE of 0.186, RMSE of 0.431 and a R2 value of 0.818. The comparison between experimental and predicted values obtained using the RFR model is illustrated in Fig. 8, indicating good agreement between the predicted and experimental values within the investigated dataset. Actual versus predicted values of compressive, split tensile and flexural strength as per RFR. Gradient boosting regressor (GBR) Gradient Boosting Regression (GBR) is another ensemble learning algorithm that sequentially combines multiple weak learners to minimize prediction error and improve model accuracy. The GBR model was trained using the same dataset and input variables used for the RFR model. The GBR model demonstrated superior predictive performance compared to the RFR model. For compressive strength prediction, the model achieved an MSE of 0.84, RMSE of 0.92, and an R2 value of 0.96, indicating a promising predictive capability within the investigated dataset. For split tensile strength prediction, the GBR model produced an MSE of 0.04, RMSE of 0.20, and an R2 value of 0.89. Similarly, for flexural strength prediction, the model achieved an MSE of 0.05, RMSE of 0.21, and an R2 value of 0.94. The relationship between predicted and experimental values for the GBR model is presented in Fig. 9. Actual versus predicted values of compressive, split tensile and flexural strength as per GBR. Model comparison and discussion Table 3 presents a comparison of the statistical performance metrics obtained for both machine learning models. Table 3Comparison of the statistical performance metrics of the machine learning models.ModelPropertyMSERMSER2RFRCompressive strength1.6881.2990.934RFRSplit tensile strength0.0780.2800.866RFRFlexural strength0.1860.4310.818GBRCompressive strength0.840.920.96GBRSplit tensile strength0.040.200.89GBRFlexural strength0.050.210.94 Comparison of the statistical performance metrics of the machine learning models. A comparison of the prediction performance of the two models indicates that the Gradient Boosting Regression model outperforms the random forest model for predicting the mechanical properties of RAP-based fibre reinforced concrete. The GBR model consistently produced lower prediction errors and higher R2 values for all three strength parameters. From a mechanistic perspective, RAP replacement level and steel fibre dosage were identified as the dominant input variables governing the predicted mechanical properties. RAP content primarily influenced the reduction in strength due to weaker asphalt-coated aggregate–paste bonding, whereas steel fibre dosage strongly contributed to improvements in tensile and flexural performance through crack-bridging and energy absorption mechanisms. The machine learning models successfully captured these experimentally observed material trends. The high predictive accuracy obtained from both ensemble learning models demonstrates their capability to effectively capture the nonlinear relationships between mix design parameters and the mechanical properties of RAP-based concrete. Furthermore, the ML models successfully reproduced the experimental trends observed in this study, including the reduction in mechanical strength with increasing RAP content and the improvement in tensile and flexural performance due to steel fibre reinforcement. These results highlight the potential of machine learning approaches for predicting the performance of sustainable concrete materials using limited experimental datasets. Such predictive models can support data-driven optimization of RAP-based concrete mixtures and reduce the need for extensive experimental testing in future studies. Although the ensemble learning models demonstrated satisfactory prediction performance, the limited dataset size may influence the generalization capability of the developed models. Therefore, the machine learning results should be interpreted as preliminary predictive models applicable within the investigated experimental range. Stress–strain behaviour of RAP-based concrete Experimental stress–strain response The compressive stress–strain behaviour of RAP-based concrete was evaluated to understand the influence of RAP aggregates and steel fibre reinforcement on the deformation characteristics of the material. Stress–strain curves were obtained from compression tests on cylindrical specimens, and the average response of three specimens was considered for each mixture. The resulting curves provide insight into the stiffness, peak stress, and post-peak behaviour of RAP-based concrete with varying RAP content and fibre dosage. The stress–strain curves for concrete mixtures incorporating different RAP replacement levels are presented in Fig. 10. The RAP-based concrete without fibres displayed a brittle failure response comparable to that of natural aggregate concrete. However, increasing steel fibre dosage enhanced post-peak toughness and ductility, mitigating brittle failure. The addition of 1.25% steel fibres led to fibre balling, which not only reduced fibre dispersion efficiency but also adversely impacted the stress–strain response of the concrete. Furthermore, the stress-ratio and strain-ratio analysis in Table 4 confirms that steel fibre incorporation significantly improves both peak-stress and peak-strain, reinforcing the structural integrity and load-bearing capacity of RAP-based concrete. Specifically, the addition of 0.5%, 0.75%, and 1% fibres resulted in an average increase of 2%, 6%, and 10% in peak-stress, respectively. Meanwhile, the peak-strain exhibited a more substantial improvement, with average increments of 14%, 22%, and 48% for the corresponding fibre dosages. These results highlight the positive influence of fibre reinforcement in enhancing the load-bearing capacity and ductility of RAP-based concrete. The energy absorbed or compressive toughness of RAP-based concrete enhanced by 1.7, 2.3, and 4.9 times with the addition of 0.5, 0.75, and 1% fibres, respectively. Stress–strain curves of RAP-based concretes with and without steel fibres. Table 4Variation of peak strain, peak stress, and energy absorption of RAP-based concrete with and without steel fibres.Mix designationPeak strainPeak stress (MPa)Pre-peak toughness (kJ/m3)Total compressive toughness (kJ/m3)Toughness indexNAC/RAP00.00212040.3757.3067.77–RAP100.00221836.3253.7260.40–RAP200.00229633.9350.7256.32–RAP300.00232629.3944.4148.94–RAP0-SF0.50.00240141.0268.13107.521.59RAP10-SF0.50.00250437.1963.16102.931.70RAP20-SF0.50.00262734.9062.1798.451.75RAP30-SF0.50.00266330.1353.5985.221.74RAP0-SF0.750.00257142.3377.03156.272.31RAP10-SF0.750.00269538.6974.42145.052.40RAP20-SF0.750.00278835.9969.80130.522.32RAP30-SF0.750.00285130.9065.90112.042.29RAP0-SF10.00312244.27104.79343.155.06RAP10-SF10.00323639.5796.70299.274.96RAP20-SF10.00341937.6896.54266.014.72RAP30-SF10.00353432.4685.65244.595.00 Variation of peak strain, peak stress, and energy absorption of RAP-based concrete with and without steel fibres. Stress–strain curve modelling Although the experimental stress–strain curves provide important insight into the compressive behaviour of RAP-based concrete, analytical modelling is required to represent the stress–strain relationship in a generalized form. Such models are useful for predicting deformation behaviour and for incorporating material response into structural analysis and numerical simulations. The objective of the proposed analytical model is to provide a simplified representation of the experimentally observed stress–strain behaviour of RAP-based fibre reinforced concrete within the investigated parameter range. In this study, the experimentally obtained curves were analysed and modelled using a modified form of the Saenz equation, which is widely used to represent the compressive stress–strain behaviour of concrete due to its ability to reasonably represents both the ascending and descending branches of the stress–strain curve. Development of equation for Stress ratio and Strain ratio From the values of peak stress and peak strain from Table 4, it can be observed that these values for steel fibre reinforced RAP-based concrete depends on both RAP content and fibre content. Generally, the behaviour of FRC depends on both volume of fibres (Vf) as well as its aspect ratio (l/d). Reinforcing Index (RI), the product of Vf and l/d is used to integrate these two factors into one term. To find the effect of RAP on stress ratio and strain ratio, the variation of normalized stress ratio and normalized strain ratio were plotted (Fig. 11a,b), and equations for variation of stress ratio and strain ratio with RAP content (RAP) were established. Variation of (a) normalized stress ratio and (b) normalized strain ratio with RAP content. Similarly, to find the effect of fibres on stress ratio and strain ratio, the variation of normalized stress ratio and normalized strain ratio were plotted (Fig. 12a,b), and equations for variation of stress ratio and strain ratio with RI were established. Variation of (a) normalized stress ratio and (b) normalized strain ratio with Reinforcing Index. Combining Eqs. (1), (3), and Eqs. (2), (4), equations for variation of stress ratio, and strain ratio with both RAP content (R) and RI were subsequently established. Table 5 shows the experimental and analytical values of stress ratio and strain ratio. It can be noted that the analytical values are in acceptable agreement within the investigated experimental range. Table 5Experimental and analytical values of stress ratio and strain ratio of RAP-based concretes with and without steel fibres.Reinforcing index (RI)RAP content (RAP)ExperimentalAnalyticalStress ratioStrain ratioStress ratioStrain ratio001.0001.0001.0001.0000.10.9001.0460.9121.0360.20.8401.0830.8241.0720.30.7281.0970.7361.1080.2501.0161.1331.0241.1080.10.9211.1810.9341.1480.20.8651.2390.8441.1880.30.7461.2560.7541.2270.37501.0491.2131.0561.2570.10.9581.2710.9631.3030.20.8921.3150.8701.3480.30.7651.3450.7771.3930.501.0971.4731.1011.4710.10.9801.5261.0041.5230.20.9331.6130.9071.5760.30.8041.6670.8101.629 Experimental and analytical values of stress ratio and strain ratio of RAP-based concretes with and without steel fibres. Normalized stress–strain model development The stress–strain curves in Fig. 10 for RAP-based concrete with different fibre content shows identical structural behaviour. Therefore, to better understand the stress–strain behaviour of RAP-based concretes, normalized stress–strain curves are developed. Normalized stress–strain curves are obtained by dividing stress and strain values at a given load level by their respective peak stress and strain values, resulting in a dimensionless representation of the stress–strain behaviour. This normalization process effectively eliminates the influence of concrete strength, allowing for a more generalized comparison of different mixes. Figure 13 presents the normalized stress–strain curves at various stress levels, highlighting that while these curves are independent of peak stress and peak strain, they still exhibit variability based on RAP content and fibre dosage. Therefore, to accurately represent the mechanical behaviour of RAP-based concrete with varying fibre dosages, it is crucial to develop a unified equation that can effectively model the stress–strain response across different RAP and fibre contents. The general equation for the compressive stress–strain behaviour of concrete specimens proposed by Saenz75 is Normalized stress–strain curves of RAP-based concretes with and without steel fibres. where ‘ε’ is the strain and ‘f’ is the stress in the concrete specimens. To determine the normalized stress–strain curve of fibre reinforced RAP-based concrete, instead of stress and strain, strain ratio and stress ratio are taken as independent and dependent variables. The following modified Saenz equation is used for developing the normalized stress–strain curve. where f/fu and ε/εu are the stress and strain ratios, respectively. ‘a’, ‘b’, and ‘c’ are constants that vary with RAP content and fibre content. Two distinct sets of constants (‘a’, ‘b’, and ‘c’) are proposed to characterize the ascending and descending portions of the stress–strain curve. To accurately determine these constants, appropriate boundary conditions must be established for both segments of the normalized stress–strain curve, ensuring a precise mathematical representation of the material’s mechanical response under loading and post-peak behaviour. For both ascending and descending portions, the common boundary conditions are The additional boundary condition for ascending and descending portions of the normalized stress–strain curve for different RAP contents and fibre contents are as shown in Table 6. The constants ‘a’, ‘b’, and ‘c’, satisfying these boundary conditions for ascending and descending portions of the normalized stress–strain curve for different RAP contents and fibre contents, are given in Table 7. From Table 7, it can be observed that, for both ascending and descending portions, the values of ‘a’, ‘b’ are inter-dependent, whereas, the value of ‘c’ is 1. The constant ‘a’ in ascending portion (‘aasc’) depends on RAP content (RAP), and fibre content, whereas the constant ‘a’ in descending portion, (‘adsc’) depends on only fibre content. Table 6Additional boundary conditions for RAP-based concretes with and without steel fibres.Reinforcing index (RI)RAP content (RAP)Ascending portionDescending portion00ε/εu. = 0.647, f/fu . = 0.85ε/εu. = 1.135, f/fu. = 0.850.1ε/εu. = 0.698, f/fu. = 0.85ε/εu. = 1.092, f/fu. = 0.850.2ε/εu = 0.740, f/fu = 0.85ε/εu = 1.079, f/fu = 0.850.3ε/εu = 0.756, f/fu = 0.85ε/εu. = 1.071, f/fu. = 0.850.250ε/εu = 0.607, f/fu = 0.85ε/εu = 1.161, f/fu = 0.85ε/εu = 0.655, f/fu = 0.85ε/εu = 184, f/fu. = 0.850.2ε/εu = 0.700, f/fu = 0.85ε/εu = 1.154, f/fu = 0.850.3ε/εu = 0.722, f/fu = 0.85ε/εu96, f/fu = 0.850.3750ε/εu = 0.562, f/fu = 0.85ε/εu = 1.272, f/fu = 0.850.1ε/εu = 0.612, f/fu = 0.85ε/εu = 1.256, f/fu = 0.850.2ε/εu = 0.658, f/fu = 0.85ε/εu = 1.186, f/fu = 0.850.3ε/εu = 0.686, f/fu = 0.85ε/εu = 1.161, f/fu = 0.850.50ε/εu = 0.476, f/fu = 0.85ε/εu = 1.416, f/fu = 0.850.1ε/εu = 0.536, f/fu = 0.85ε/εu = 1.382, f/fu = 0.850.2ε/εu = 0.590, f/fu = 0.85ε/εu = 1.361, f/fu = 0.850.3ε/εu = 0.622, f/fu = 0.85ε/εu = 1.373, f/fu = 0.85 Additional boundary conditions for RAP-based concretes with and without steel fibres. Table 7Constants for ascending and descending portions of stress–strain curve for RAP-based concretes with and without steel fibres.Reinforcing index (RI)RAP content (RAP)Ascending portionDescending portionaascbasccascadscbdsccdsc001.090 − 0.91010.091− 1.90910.10.738− 1.26210.044− 1.95610.20.518− 1.48210.033− 1.96710.30.447− 1.55310.026− 1.97410.2501.444− 0.55610.126− 1.87410.11.039− 0.96110.162− 1.83810.20.729− 1.27110.117− 1.88310.30.605− 1.39510.182− 1.81810.37501.934− 0.06610.329− 1.67110.11.399− 0.60110.295− 1.70510.21.006− 0.99410.165− 1.83510.30.814− 1.18610.127− 1.87310.503.2731.27310.694− 1.30610.12.2700.27010.599− 1.40110.21.615− 0.38510.543− 1.45710.31.302− 0.69810.575− 1.4251 Constants for ascending and descending portions of stress–strain curve for RAP-based concretes with and without steel fibres. The variation of ‘aasc’ with RAP content for different fibre contents / RI is plotted in Fig. 14. It can be observed that the variation of ‘aasc’ with RAP content is almost similar, and a generalized equation may be proposed for variation of ‘aasc’ with RAP content that is independent of fibre content. Figure 15 shows the variation of normalized ‘aasc’ with RAP content. The variation of normalized aasc can be written as Variation of aasc with RAP content for different fibre contents. Variation of (aasc)N with RAP content. where, or, and, ‘a0’ is the value of ‘aasc’ at RAP content of 0 for different values of RI. Figure 16 shows the variation of ‘a0’ with RI2, and the variation of ‘a0’ with RI can be written as Variation of a0 with RI2. Substituting the Eqs. (10) and (13) in Eq. (12) the following expressions can be obtained. or, The variation of ‘adsc’ with RAP content for different fibre contents / RI is almost negligible. The variation of ‘adsc’ with RI2 is plotted in Fig. 17 and the variation of ‘adsc’ with RI can be written as Variation of adsc with RI2. From Eqs. (14) and (15) values of ‘aasc’ and ‘adsc’ are calculated and listed in Table 8. Since the value of ‘b’ is dependent on values of ‘a’, ‘basc’, and ‘bdsc’ are calculated from values of‘aasc’ and ‘adsc’ and listed in Table 8. Table 8Analytical constants for ascending and descending portions of stress–strain curve for RAP-based concretes with and without steel fibres.Reinforcing index (RI)RAP content (RAP)Ascending portionDescending portionaascbasccascadscbdsccdsc001.090− 0.91010.049− 1.95110.10.795− 1.20510.049− 1.95110.20.581− 1.41910.049− 1.95110.30.424− 1.57610.049− 1.95110.2501.429− 0.57110.095− 1.90510.11.043− 0.95710.095− 1.90510.20.761− 1.23910.095− 1.90510.30.555− 1.44510.095− 1.90510.37502.0040.00410.218− 1.78210.11.462− 0.53810.218− 1.78210.21.067− 0.93310.218− 1.78210.30.779− 1.22110.218− 1.78210.503.2181.21810.694− 1.30610.12.3480.34810.694− 1.30610.21.714− 0.28610.694− 1.30610.31.251− 0.74910.694− 1.3061 Analytical constants for ascending and descending portions of stress–strain curve for RAP-based concretes with and without steel fibres. Model validation and comparison Using the analytical constants obtained from the modified Saenz model, normalized analytical stress–strain curves were generated, as shown in Fig. 18. These normalized curves were subsequently converted into actual stress–strain relationships by incorporating the analytical stress ratio and strain ratio obtained in the previous sections. The resulting analytical stress–strain curves are presented in Fig. 19. A comparison between the analytical and experimental stress–strain curves for varying RAP contents at a fixed fibre dosage of 1% is illustrated in Fig. 20, while Fig. 21 presents a comparison for different fibre dosages at a constant RAP content of 20%. The close agreement between the analytical predictions and experimental results confirms that the proposed model reasonably represents the compressive behaviour of RAP-based fibre reinforced concrete. The model effectively represents both the ascending and post-peak portions of the stress–strain curve and accounts for the influence of RAP content and fibre reinforcement. To quantitatively evaluate the predictive capability of the proposed stress–strain model, the coefficient of determination (R2) between the analytical and experimental stress–strain responses was evaluated for all investigated mixtures. The obtained R2 values ranged between approximately 0.93 and 0.98, indicating satisfactory agreement between the proposed analytical model and the experimental observations within the investigated parameter range. Normalized analytical stress–strain curves of RAP-based concretes with and without steel fibres. Analytical stress–strain curves of RAP-based concretes with and without steel fibres. Analytical and Experimental stress–strain curves for different RAP contents at a fibre content of 1% Analytical and Experimental stress–strain curves for different fibre content at a RAP content of 20% Overall, the proposed analytical model provides a reliable representation of the compressive stress–strain behaviour of RAP-based concrete mixtures. The model constants were calibrated for the investigated RAP replacement levels and steel fibre dosages. Although continuous regression formulations relating the model constants to RAP content and reinforcing index may be developed using larger datasets, the present study intentionally adopts discrete calibrated coefficients to avoid overgeneralization beyond the experimentally investigated range. The improved ductility and enhanced post-peak response observed in fibre reinforced mixtures are expected to contribute positively to fatigue resistance under cyclic loading, as discussed in the following section. Flexural fatigue behaviour of RAP-based concrete Based on the strength and stress–strain behaviour of RAP-based concrete mixes, 1% steel fibre dosage was identified as the optimum content for enhancing mechanical performance. Consequently, the flexural fatigue behaviour of RAP-based concrete was evaluated exclusively at this fibre dosage. Tables 9 and 10 present the fatigue life data for various RAP proportions across different stress levels, both with and without steel fibre reinforcement. Table 9Fatigue life data of RAP-based concrete mixes without fibre.Mix designationStress level (S)Fatigue life (no. of cycles to failure)Sample 1Sample 2Sample 3NAC/RAP00.963931390.82619355344780.719,98324,41828,299RAP100.959831250.81944311240110.719,05220,87625,592RAP200.947731170.81464357247390.714,52719,04025,442RAP300.937721070.81454274842760.712,47616,68222,148 Fatigue life data of RAP-based concrete mixes without fibre. Table 10Fatigue life data of RAP-based concrete mixes with fibre.Mix designationStress level (S)Fatigue life (no. of cycles to failure)Sample 1Sample 2Sample 3RAP0-1%SF0.92944857180.812,30818,45826,3710.7138,195171,340190,508RAP10-1%SF0.92053695690.813,31918,71028,3480.7131,122141,025195,251RAP20-1%SF0.91753886760.812,69119,00421,4200.7115,161138,025162,175RAP30-1%SF0.91402704670.8700514,40425,2470.798,052115,825135,252 Fatigue life data of RAP-based concrete mixes with fibre. The fatigue life of RAP-based concrete mixtures shows a gradual reduction with increasing RAP content. This may be due to the increased heterogeneity of concrete matrix and the presence of aged binder that may be attributed to weaker interfacial transition zone characteristics when RAP aggregates are used. These weak zones act as preferential sites for crack initiation under fatigue conditions. As a result, the inclusion of RAP leads to strain localization, particularly at the RAP-cement interface, leading to earlier fatigue crack formation. However, the incorporation of steel fibres significantly improves the fatigue performance of RAP-based concrete. The randomly distributed fibres bridge micro-cracks and transfer stresses across crack surfaces through fibre pull-out mechanisms, thereby delaying crack propagation and enhancing the energy absorption capacity of the material. As a result, fibre reinforced RAP-based concrete exhibits substantially higher fatigue life compared with plain RAP-based concrete mixtures. The fibres also help redistribute stress concentrations caused by material heterogeneity, resulting in more stable crack development and improved fatigue resistance. The fatigue behaviour observed in the present study is consistent with the general trend reported for fibre reinforced concrete, where steel fibres contribute to delaying crack initiation and crack propagation under repeated loading. Fatigue life data exhibited the inherent scatter commonly associated with cyclic loading behaviour of concrete materials. The coefficient of variation (COV) of fatigue life results remained within acceptable experimental limits for fatigue testing of heterogeneous cementitious composites. The observed variability is primarily attributed to the stochastic nature of micro-crack initiation and propagation under repeated loading. The S–N curve (fatigue equation) is widely used to characterize fatigue behaviour in concrete, and Figs. 22 and 23 present the S–N curves and corresponding fatigue equations for different concrete mixes, both with and without fibres. These curves illustrate the expected trend in which the fatigue life increases as the applied stress level decreases. The regression equations derived from the S–N curves describe the relationship between stress ratio and fatigue life for each concrete mixture. The coefficient of determination (R2) values for the developed regression models ranges between 0.94 and 0.97, indicating strong agreement between the experimental data and the regression equations. S–N curves for RAP-based concrete without fibres. S–N curves for RAP-based concrete with steel fibres. The endurance limit defined as the minimum stress ratio at which the specimen can bear 2 million load cycles was calculated by means of these S–N equations. The S–N equations and the calculated endurance limits for the different concrete mixtures are summarized in Table 11. For RAP-based concrete without fibre reinforcement the endurance limit is 0.544–0.561. The slight decrease in endurance limit with the increase of RAP content can be explained by the decreased strength of the bond between RAP aggregates and cementitious matrix. It should also be noted that the endurance limit values corresponding to 2 × 106 cycles were obtained through logarithmic extrapolation of the experimentally derived S–N relationships. Since the experimental fatigue data extended up to approximately 105–106 loading cycles, the predicted endurance limits should therefore be interpreted as analytical estimates within the adopted regression framework. Table 11S–N Equations and Endurance limit for RAP-based concrete.Mix designationRegression equation/S–N equationR2Endurance limitNAC/RAP0S = 1.0626–0.0346ln(N)0.960.561RAP10S = 1.0596–0.0348ln(N)0.960.555RAP20S = 1.0497–0.0344ln(N)0.940.551RAP30S = 1.0464–0.0340ln(N)0.940.544RAP0-1%SFS = 1.1071–0.0329ln(N)0.970.630RAP10-1%SFS = 1.0989–0.0324ln(N)0.950.629RAP20-1%SFS = 1.0926–0.0320ln(N)0.950.628RAP30-1%SFS = 1.0763–0.0310ln(N)0.940.627 S–N Equations and Endurance limit for RAP-based concrete. In comparison, the addition of steel fibres greatly enhances the fatigue resistance. Steel fibre reinforced RAP-based concrete has endurance limits ranging from 0.630 to 0.637 which is an improvement of around 12–15% as compared with concrete mixtures with no fibres. Figure 24 shows the variation in endurance limit with RAP contents for plain and fibre reinforced concrete mixtures. Although because of the RAP inclusion there is a marginal decrease in endurance limit, with the inclusion of the steel fibres the overall fatigue strength of the material is improved significantly. Endurance limit variation of RAP-based concretes with and without steel fibres. The better fatigue performance of fibre reinforced RAP-based concrete may be attributed to the fibre-bridging mechanisms of steel fibres leading to delayed crack initiation, reduced crack propagation rates and enhanced energy absorption capacity of concrete under cyclic loading which is commonly reported for fibre reinforced concrete. Consequently, fibre reinforcement has a strong effect on fatigue durability of RAP-based concrete and makes it more suitable for pavement applications where it is subjected to repeated traffic loading. Although some variability in fatigue life was observed, the overall fatigue trends remained consistent across the investigated stress levels and fibre dosages. Overall, the fatigue analysis reveals that although the incorporation of RAP causes slightly decreased fatigue life because of the weaker interface bonding, the introduction of steel fibres is well able to overcome these shortcomings while considerably improving fatigue resistance. These results highlight the potential of steel fibre reinforced RAP-based concrete as a sustainable and durable material for pavement applications. The fatigue relationships developed in the present investigation are limited to the investigated stress levels, material combinations, and loading conditions. The present study focused on deterministic fatigue behaviour within selected stress levels commonly adopted for pavement concrete evaluation. Probabilistic fatigue analysis and reliability-based assessment were beyond the scope of the present investigation and may be considered in future studies involving broader experimental datasets and loading conditions. Limitations and future scope The present investigation was conducted using a specific RAP source, one concrete grade, selected RAP replacement levels, and steel fibre dosages under controlled laboratory conditions. Accordingly, the developed analytical stress–strain model and machine learning models were calibrated using the experimental data generated in the present study and are therefore applicable within the investigated parameter range. Although the proposed stress–strain model reasonably represents the experimentally observed behaviour, broader validation using independent datasets, different RAP sources, varying concrete grades, and wider material combinations is necessary before wider generalization. In addition, the fatigue evaluation was limited to selected stress levels representing typical pavement loading conditions. Direct microstructural characterization techniques such as SEM, XRD, or CT imaging were not included within the scope of the investigation; therefore, the mechanistic interpretations related to interfacial transition zone behaviour, asphalt-coated RAP interaction, and steel fibre bridging mechanisms were based primarily on experimentally observed trends and findings reported in previous literature. Furthermore, the statistical interpretation and predictive capability presented in this study are dependent on the experimental repeatability and dataset size considered in the investigation, and may be further strengthened through larger datasets and multi-source validation studies. Future research should therefore focus on durability performance, long-term fatigue behaviour, advanced microstructural characterization, hybrid fibre systems, varying environmental exposure conditions, and large-scale validation of the developed analytical and machine learning models for broader practical implementation. Conclusions Based on the experimental investigation, stress–strain analysis, fatigue evaluation, and machine learning modelling of RAP-based concrete with steel fibre reinforcement, the following conclusions may be drawn: The incorporation of reclaimed asphalt pavement (RAP) aggregates results in a gradual reduction in the mechanical strength of concrete, which may be attributed to relatively weaker interfacial characteristics between the asphalt-coated RAP particles and the cementitious matrix. However, concrete containing up to 20% RAP satisfies the 40 MPa compressive strength requirement, while mixes containing 30% RAP meet the flexural strength requirement for Pavement Quality Concrete (PQC) as per MoRTH specifications. The incorporation of reclaimed asphalt pavement (RAP) aggregates results in a gradual reduction in the mechanical strength of concrete, which may be attributed to relatively weaker interfacial characteristics between the asphalt-coated RAP particles and the cementitious matrix. However, concrete containing up to 20% RAP satisfies the 40 MPa compressive strength requirement, while mixes containing 30% RAP meet the flexural strength requirement for Pavement Quality Concrete (PQC) as per MoRTH specifications. The incorporation of steel fibres improved the tensile and flexural performance of RAP-based concrete, particularly at fibre dosages up to 1%. The observed enhancement in performance is consistent with crack-bridging and crack-arresting mechanisms commonly reported for fibre reinforced concrete. Beyond 1% fibre dosage, fibre agglomeration reduced fibre dispersion efficiency and marginally affected the mechanical performance. The incorporation of steel fibres improved the tensile and flexural performance of RAP-based concrete, particularly at fibre dosages up to 1%. The observed enhancement in performance is consistent with crack-bridging and crack-arresting mechanisms commonly reported for fibre reinforced concrete. Beyond 1% fibre dosage, fibre agglomeration reduced fibre dispersion efficiency and marginally affected the mechanical performance. Machine learning models based on Random Forest Regression and Gradient Boosting Regression were developed to predict the mechanical properties of RAP-based concrete mixtures within the investigated dataset. Among the considered models, the Gradient Boosting model demonstrated comparatively better predictive capability for estimating the mechanical properties of RAP-based concrete mixtures within the investigated experimental range. Machine learning models based on Random Forest Regression and Gradient Boosting Regression were developed to predict the mechanical properties of RAP-based concrete mixtures within the investigated dataset. Among the considered models, the Gradient Boosting model demonstrated comparatively better predictive capability for estimating the mechanical properties of RAP-based concrete mixtures within the investigated experimental range. Steel fibre reinforcement improved the deformation characteristics of RAP-based concrete by enhancing both peak stress and peak strain. In addition, the compressive toughness increased significantly with fibre incorporation, indicating improved ductility and energy absorption capacity of the concrete mixtures. Steel fibre reinforcement improved the deformation characteristics of RAP-based concrete by enhancing both peak stress and peak strain. In addition, the compressive toughness increased significantly with fibre incorporation, indicating improved ductility and energy absorption capacity of the concrete mixtures. A stress–strain model based on the modified Saenz equation was successfully developed to represent the compressive behaviour of steel fibre reinforced RAP-based concrete. The proposed model reasonably represented both the ascending and descending branches of the experimentally observed stress–strain response within the investigated RAP replacement levels and fibre dosages. A stress–strain model based on the modified Saenz equation was successfully developed to represent the compressive behaviour of steel fibre reinforced RAP-based concrete. The proposed model reasonably represented both the ascending and descending branches of the experimentally observed stress–strain response within the investigated RAP replacement levels and fibre dosages. The fatigue performance of RAP-based concrete showed a gradual reduction with increasing RAP content, which may be associated with reduced interfacial bonding characteristics. However, the addition of steel fibres is highly beneficial for fatigue performance by postponing crack initiation and retarding crack propagation in cyclic loading. The fatigue performance of RAP-based concrete showed a gradual reduction with increasing RAP content, which may be associated with reduced interfacial bonding characteristics. However, the addition of steel fibres is highly beneficial for fatigue performance by postponing crack initiation and retarding crack propagation in cyclic loading. Steel fibre reinforcement enhanced the endurance limit by about 12–15% proving its effectiveness in enhancing the fatigue resistance and durability of RAP-based concrete under repeated loading regimes. Steel fibre reinforcement enhanced the endurance limit by about 12–15% proving its effectiveness in enhancing the fatigue resistance and durability of RAP-based concrete under repeated loading regimes. Overall, the findings of the present study indicate the potential applicability of steel fibre reinforced RAP-based concrete for sustainable pavement-related applications through the utilization of recycled pavement materials while maintaining acceptable mechanical and fatigue performance within the investigated experimental conditions. Future research may focus on the long-term durability performance of steel fibre reinforced RAP-based concrete, including shrinkage, permeability, and environmental exposure conditions. In addition, large-scale pavement trials and sophisticated data-driven modelling methods could continue to validate and optimise the practical application of RAP-based fibre reinforced concrete to sustainable pavement infrastructure. Acknowledgements The authors gratefully acknowledge the Department of Civil Engineering, National Institute of Technology (NIT) Warangal, for providing the necessary facilities and support to carry out the experimental investigations presented in this study. The authors sincerely thank the faculty and technical staff at NIT Warangal for their assistance during experimental study. Author contributions Y.P. and S.R.R.T.P. wrote the main manuscript text. Y.P. and S.S.A.B.P. prepared the figures and handled visualization. Data curation was carried out by Y.P. and S.S.A.B.P., while formal analysis was performed by S.R.R.T.P. Investigation and methodology were contributed by Y.P., S.R.R.T.P., and A.M.K. Conceptualization and supervision were provided by H.M. and R.K.P. All authors reviewed and approved the final manuscript. Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Data availability Data used for the analysis is provided within the manuscript. 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. References Associated Data Data Availability Statement Data used for the analysis is provided within the manuscript.