Accurate forecasting of agricultural crop prices is essential for enabling informed decision-making among farmers and market stakeholders, particularly in regions characterized by high price volatility. This paper presents a hybrid machine learning framework that integrates AutoRegressive Integrated Moving Average with exogenous variables (ARIMAX) and Long Short-Term Memory (LSTM) networks for 90-day crop price forecasting. The proposed model captures linear and seasonal patterns through ARIMAX while modeling non-linear residual dynamics using LSTM. The system incorporates exogenous factors such as weather conditions and policy indicators, including Minimum Support Price (MSP), to enhance predictive performance. Experiments are conducted on multiple crops from the Coimbatore district agricultural market using a chronological train-test split. The hybrid approach is evaluated against baseline models, including ARIMA, standalone LSTM, and Gradient Boosting, using standard metrics such as RMSE, MAE, and MAPE. Results demonstrate that the proposed framework achieves competitive accuracy, particularly for crops with sufficient historical data, highlighting its potential for practical deployment in agricultural decision support systems.

Time-series forecasting ARIMAX LSTM Hybrid models Agricultural price prediction Crop price forecasting Machine learning