Interpretable Crop Selection for Optimized Farming Decisions

Abstract

Agricultural success hinges on strategic crop selection, directly influencing yield, financial stability, and risk management for farmers. Despite integrating machine learning techniques, many current systems function as opaque ”black boxes,” leading to reluctance among farmers who need both precision and transparency in crop recommendations. This study introduces a novel, interpretable approach for crop selection using climate and soil data, employing the AdaBoost classifier, renowned for its high accuracy and ability to prioritize misclassified data points. To enhance transparency and foster trust among farmers, we incorporate SHapley Additive Explanations (SHAP) to elucidate the model’s decision-making process. Our system analyzes diverse parameters such as nitrogen, phosphorus, potassium, pH, temperature, humidity, and rainfall to suggest suitable crops for cultivation. Evaluated on a comprehensive dataset of 22 crops, our approach achieves exceptional accuracy (99.77%) compared to conventional and boosted models, with rapid processing times (0.5 seconds per prediction). SHAP interpretations clarify the impacts of various climate and soil factors on crop suitability, offering farmers clear justifications for the recommendations provided. By combining accuracy with transparency, our system empowers farmers to make informed decisions about their land, leading to improved yields and increased profitability. This interpretable system represents a significant advancement in developing efficient and reliable AI tools for sustainable crop selection in agriculture. We envision a future where farmers can embrace AI-driven tools with confidence, fostering a more sustainable agricultural landscape.