ML Framework for Profitable Crop Yield Based on Climatic and Economic Factors

Abstract

Agricultural planning is inherently uncertain: soil composition varies across seasons, and market prices for produce are rarely predictable at the time planting decisions are made. Farmers routinely face this uncertainty with limited analytical support, which motivates the development of the Agricultural Decision Support System (ADSS) presented in this work. The ADSS assists farmers by combining agronomic and economic data to generate crop recommendations, evaluating nitrogen, phosphorus, and potassium content, temperature, humidity, rainfall, and soil pH to identify crops suited to given field conditions and likely to be commercially viable. Crop recommendations are generated using an XGBoost classifier, achieving 98.86% accuracy across 22 crop categories. In addition to crop suitability, the system estimates expected profitability for the recommended crop; factors such as adverse weather and pest incidence remain outside the scope of the model. A deep learning autoencoder is incorporated to denoise the input feature space while preserving informative structure, improving downstream model efficiency. The resulting system operates with sub-millisecond inference latency, making it suitable for near-real-time deployment. Recommendations are accompanied by natural-language explanations, allowing users to understand the reasoning behind each suggestion. Overall, the ADSS functions as a data-driven advisory tool that jointly considers agronomic suitability and economic outcome, positioning it as a practical component of precision agriculture powered by machine learning and artificial intelligence