International Journal of Advanced Research in Computer and Communication Engineering

Abstract: Agricultural commodity prices in India's Vidarbha region (Maharashtra)—a key producer of cotton, soybean, oranges, and pulses—exhibit extreme volatility due to erratic monsoons, seasonal supply-demand imbalances, transport logistics, limited storage, and government policies like minimum support prices (MSP). This unpredictability causes financial distress for over 1.5 million farming households, traders, and consumers, who lack reliable forecasting tools beyond rudimentary historical averages or linear statistical models. Such traditional approaches fail to model the non-linear, multifaceted patterns in price time series, especially during events like droughts or market surges.

This research addresses the gap by developing and evaluating machine learning (ML) models for accurate price forecasting using historical data (2015–2024) from APMC mandis in Nagpur, Akola, and Yavatmal. After rigorous preprocessing (outlier removal, normalization, and feature engineering with lags, weather, and arrivals), we trained regression models—linear regression, support vector regression (SVR), random forest, XGBoost, and LSTM—on chronologically split datasets.

XGBoost emerged superior (test MAPE: 4.8%, R²: 0.92 for soybean), outperforming ARIMA by 60% and capturing Vidarbha-specific volatilities. LSTM excelled in long-term dependencies. These results validate ML's potential for nonlinear time series analytics, providing farmers actionable predictions for crop planning, harvest timing, storage, and sales.
Deployable via mobile apps with APMC APIs, this framework enhances decision-making, stabilizes incomes, and supports policy in climate-vulnerable regions. Future enhancements include real-time satellite integration.

Keywords: Vidarbha Agriculture, Price Forecasting, Machine Learning, XGBoost, Time Series, Commodity Volatility.

Downloads: | DOI: 10.17148/IJARCCE.2026.15211