Time-Series Forecasting of Agricultural Market Prices Using Attention-Enhanced LSTM with Weather-Informed Crop Recommendation via XGBoost

Fluctuations in prices in agricultural markets offer farmers a challenge in determining when and where to market their products, and in most cases, their incomes become uncertain. These price variabilities arise from a combination of seasonal effects, regional demand–supply imbalances, short-term weather conditions, and policy-influenced factors. Current forecasting techniques that rely heavily on pre-existing statistical models or manual methods are often unable to represent such complex and dynamic interactions with the required accuracy.This paper introduces an artificial intelligence–based model for agricultural crop price prediction that combines an Attention-Enhanced Long Short-Term Memory (LSTM) network with the XGBoost machine learning model. The models are trained using historical market price data acquired from the Agmarknet portal, enabling them to learn both temporal patterns and non-linear correlations among commodity prices. To strengthen contextual awareness, limited yet significant weather forecasting data are incorporated using a free five-day weather API, considering real-world data availability constraints.The proposed system provides localized price forecasts through a web-based dashboard supported by visual trend analysis and automated price notifications, assisting farmers in making informed selling decisions. Although the current implementation does not include mandi selection recommendations, the framework is extensible and can be refined by integrating richer meteorological data and additional market intelligence. Experimental evaluation demonstrates improved predictive performance compared to conventional forecasting methods, as indicated by lower prediction error metrics.