AgroXAI: Explainable AI-Driven Crop Recommendation System for Agriculture 4.0

Özlem TURGUT, İbrahim KÖK, Suat ÖZDEMİR · 2024

Read AgroXAI: Explainable AI-Driven Crop Recommendation System for Agriculture 4.0 by Özlem TURGUT, İbrahim KÖK, Suat ÖZDEMİR (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-16196.txt.

Introduction

Turgut, Kök, and Özdemir built this crop adviser in twenty twenty-four for smart farming. Choosing what to plant grows harder as land shrinks and climate shifts. Their system, AgroXAI, reads weather and soil data from field sensors. A machine learning model then suggests crops suited to that region.

What sets the system apart is explanation. Standard tools show which factors drove each suggestion, locally and globally. A counterfactual tool goes further: it names alternative crops and what would need to change to favor them. The design runs on edge hardware near the farm. The authors frame it as a platform for regional crop diversity in the next farming generation.

Catalog record

CREATORÖzlem TURGUT, İbrahim KÖK, Suat ÖZDEMİR
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-16196

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

AgroXAI: Explainable AI-Driven Crop Recommendation System for Agriculture 4.0 Özlem TURGUT, İbrahim KÖK, Suat ÖZDEMİR Abstract Today, crop diversification in agriculture is a critical issue to meet the increasing demand for food and to improve food safety and quality. This issue is considered to be the most important challenge for the next generation of agriculture due to diminishing natural resources, limited arable land and unpredictable climatic conditions caused by climate change. In this paper, we employ emerging technologies such as the Internet of Things (IoT), machine learning (ML) and explainable artificial intelligence (XAI) to improve operational efficiency and productivity in the agricultural sector. Specifically, we propose an edge computing-based explainable crop recommendation system, AgroXAI, which suggests suitable crops for a region based on weather and soil conditions. In this system, we provide local and global explanations of ML model decisions with methods such as ELI5, LIME, SHAP, which we integrate into ML models. More importantly, we provide regional alternative crop recommendations with the Counterfactual explainability method. In this way, we envision that our proposed AgroXAI system will be a platform that provides regional crop diversity in the next generation agriculture. I Introduction Today, global climate change, population growth, agricultural land depletion and biodiversity loss pose serious environmental, social and economic threats [ 1 ] . Under these globally changing conditions, industrial growth and production methods based on sustainable agricultural principles are needed to meet the increasing demand for food and improve food security and quality [ 2 ] . At this point, ensuring crop diversity in agriculture stands out as a critical issue. Crop diversification is the process of adding a new crop, a different variety, to a field [ 3 ] . It is not only mitigates various agricultural challenges such as soil degradation, pest infestation, and climate change impacts, but also enhances farm resilience, fosters income growth on small holdings, stabilizes commodity prices, improves soil health, and offers a diverse range of nutritious food options for both humans and livestock [ 4 ] . In this context, ensuring regional crop diversity based on low-cost digital technologies, in response to changing environmental and climate conditions, emerges as a key area of focus for agriculture 4.0 [ 5 ] . In particular, it is necessary to periodically monitor the soil, air, and environmental factors in distributed geographic regions and provide suitable crop alternatives accordingly. The concept of the Internet of Things (IoT), which provides technology-oriented creative solutions to the current needs in the new agricultural era, comes to the forefront [ 6 , 7 ] . IoT offers innovative digital solutions in many areas based on low-cost, low-latency approaches such as edge and fog computing that respond to end-user needs on-site [ 8 ] . Many of these solutions are also being used effectively in agriculture domain. In this context, a wide variety of IoT-supported smart agriculture applications are implemented in crop management, soil management, water management, livestock management, green management, weather management, and tracking and tracing [ 9 , 10 ] . It is observed that these applications are becoming widespread with AI support, but there is also a growing need for interpretability and explainability for the effective use of these technologies. In this regard, Explainable Artificial Intelligence (XAI) plays a crucial role in agriculture by enhancing the transparency of AI-driven recommendations, such as those for crop selection, soil health, and pest management [ 11 ] . By clarifying how environmental factors, soil properties, and climate conditions shape AI suggestions, XAI builds trust in technological solutions and supports data-driven decision-making. This interpretability is essential for sustainable agricultural practices, allowing stakeholders to make informed choices that enhance resilience and productivity amid changing environmental conditions. Specifically, the existing research efforts related to crop prediction in the agriculture domain are as follows. Sharma et al. [ 12 ] developed a crop prediction model based on Gaussian Naive Bayes, Decision Tree, Logistic Regression, Random Forest, and XGBoost models. The authors developed their models by training them on data collected from different agricultural fields in India. The forecasting system based on the developed models was enabled to be used online and offline via an Android application. Similarly, Doshi et al. [ 13 ] developed an intelligent recommendation system called AgroConsultant to help farmers in India make an intelligent decision on which crop to grow depending on various environmental and geographical factors. In the developed system, the authors presented a system design that combines data sets showing soil, climate and precipitation characteristics and decides which crop farmers will grow by using Decision Tree, K-Nearest Neighbour, Random Forest, regression and Neural Network models. Rajak et al. [ 14 ] aimed to design a crop recommendation system using an ensemble learning model based on Support Vector Machine, Naive Bayes, Multi-layer Perceptron and Random Forest models. The proposed system was tested with soil dataset collected from different sources. Rule-based output was produced with the ensemble model used in the system. Similarly, Pudumalar et al. [ 15 ] aimed to design a highly accurate and effective crop recommendation system using ensemble techniques. They applied the majority voting technique in Random Forest, CHAID, K-Nearest Neighbour and Naive Bayes learners. In this study, soil testing lab and crop data available in online sources were used as a dataset. As a result of the applied model, they created a rule. Saranya et al. [ 16 ] developed a crop recommendation module that ensembles the results of Support Vector Machine, Random Forest, Naive Bayes, and K-Nearest Neighbor models using majority voting. In addition, they designed a web-based interface called TILLAGE that includes fertilizer recommendation and pesticide recommendation modules based on the CNN model. The proposed modular solution aims to increase agricultural production by improving soil use, fertilization and crop selection. Unlike the above studies, edge computing architecture and Deep Reinforcement learning (DRL) learning-based methods have been used in the architectural design and product management of smart agriculture applications. For example; Wang [ 17 ] aimed to design an edge-cloud computing-based smart agriculture system using DRL. This system consists of an agricultural data collection layer, edge computing layer, data transmission layer, and cloud computing layer. While data is collected from various sensors placed in the agricultural data collection layer, DRL is deployed in the cloud computing layer for instant smart decision-making. In another study, Alonso et al. [ 18 ] proposed an architecture called Global Edge Computing Architecture (GECA) to realize Edge-IoT applications. In this architecture, they developed SDN/NFV capabilities and a Double Deep-Q Learning model to control virtual data flows. Din et al. [ 19 ] focused on the problem of monitoring crop health in a semi-structured farm. For this problem, a non-uniform area coverage algorithm based on the Double Deep Q-Network (DDQN) algorithm was developed. On the other hand, studies on the explainability of smart agriculture practices have been quite limited. Sabrina et al. [ 20 ] designed an ML and fuzzy logic-based smart agriculture system aiming to increase crop production. The system is based on SVR, KNN and Naive Bayes models to monitor soil properties, and weather conditions, and detect possible abnormal conditions for the target crop. In the paper, an approach based on fuzzy rules is presented to ensure user

Read the complete text