International Journal of Advanced Research in Computer and Communication Engineering

Abstract: Accurate prediction of agricultural commodity prices is a critical challenge due to the highly dynamic and nonlinear nature of agricultural markets. Crop prices are influenced by multiple interdependent factors such as climatic variations, seasonal demand, production levels, supply chain disruptions, government policies, and global trade dynamics. Traditional forecasting techniques, including linear regression and classical time-series models such as ARIMA, often fail to model these complex interactions and long-term temporal dependencies, leading to limited prediction accuracy and poor adaptability under volatile market conditions.

This paper proposes a hybrid crop price prediction framework that integrates Machine Learning and Deep Learning techniques to achieve reliable and long-term agricultural price forecasting. The proposed system combines the strengths of Extreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM) networks. XGBoost is employed to effectively model structured features and nonlinear relationships among economic, seasonal, and meteorological variables, while LSTM networks are utilized to capture long-term sequential dependencies and temporal trends in historical crop price data. An ensemble strategy is applied to merge predictions from both models, enhancing robustness and reducing forecasting error.

The system utilizes historical market prices along with weather-related parameters such as temperature, rainfall, and humidity to generate price forecasts for multiple crops and market locations up to twelve months in advance. The proposed framework is implemented using a scalable web-based architecture, featuring a React-based interactive dashboard for visualization and a FastAPI-powered backend for efficient data processing and real-time prediction. Experimental evaluation using standard performance metrics including RMSE, MAE, MAPE, and R² score demonstrates that the hybrid ensemble model consistently outperforms individual machine learning and deep learning models.

Downloads: | DOI: 10.17148/IJARCCE.2026.15171