Yenkikar Anuradha, Mishra Ved Prakash, Bali Manish, Ara Tabassum
School of Engineering, Amity University Dubai Campus, Dubai, 25314, UAE.
Department of CSE (AI), Vishwakrma Institute of Technology, Pune, 411048, Maharashtra, India.
MethodsX. 2025 Jun 17;15:103442. doi: 10.1016/j.mex.2025.103442. eCollection 2025 Dec.
Agriculture is a major contributor to India's GDP and employs a large population. Key crops like rice are essential for food security, making higher yields crucial for sustainability. The use of machine learning (ML) in crop yield prediction has significantly improved forecast accuracy. However, the adoption of these models by policymakers and farmers is hindered by their lack of interpretability. Explainable Artificial Intelligence (XAI) techniques address this challenge by making AI-driven predictions more transparent, ensuring trust and better decision-making. This research integrates XAI techniques into a hybrid model that combines the powers of Random Forest (RF), Long Short-Term Memory (LSTM), and XGBoost algorithms by incorporating SHAP (SHapley Additive Explanations), LIME (Local Interpretable Model-Agnostic Explanations), and Counterfactual Analysis for yield prediction. On a large-scale, multi-year agricultural dataset comprising over 246,000 records across 33 states, spanning crops, seasons, and climatic factors provided by the Indian Agriculture Department, the model achieved high accuracy (R² = 0.9827 for crop yield and 0.9721 for rice yield) outperforming existing models. The method involves:•Implementing a hybrid AI model to improve accuracy in yield predictions.•Integrating XAI methods to enhance model transparency and interpret nuanced feature interactions•Delivering actionable insights via the developed 'E-Kisan' web interface.
农业是印度国内生产总值的主要贡献者，雇佣了大量人口。像水稻这样的主要作物对粮食安全至关重要，提高产量对可持续性发展至关重要。机器学习（ML）在作物产量预测中的应用显著提高了预测准确性。然而，政策制定者和农民对这些模型的采用受到其缺乏可解释性的阻碍。可解释人工智能（XAI）技术通过使人工智能驱动的预测更加透明，确保信任并促进更好的决策，从而应对这一挑战。本研究将XAI技术集成到一个混合模型中，该模型通过纳入SHAP（SHapley值加法解释）、LIME（局部可解释模型无关解释）和反事实分析，结合了随机森林（RF）、长短期记忆（LSTM）和XGBoost算法的能力，用于产量预测。在一个由印度农业部提供的包含33个邦超过24.6万条记录、涵盖作物、季节和气候因素的大规模多年农业数据集上，该模型实现了高精度（作物产量的R² = 0.9827，水稻产量的R² = 0.9721），优于现有模型。该方法包括：•实施一个混合人工智能模型以提高产量预测的准确性。•集成XAI方法以提高模型透明度并解释细微的特征交互作用•通过开发的“电子农民”网络界面提供可操作的见解。
