APPLICATION OF MACHINE LEARNING–ENHANCED STATISTICAL MODELS FOR PREDICTING CLIMATE-RELATED AGRICULTURAL RISKS IN PAKISTAN

Abstract

Agriculture in Pakistan is increasingly threatened by climate variability, including irregular rainfall patterns, rising temperatures, floods, and droughts, which significantly affect crop productivity and food security. Traditional statistical forecasting methods are limited in capturing the nonlinear and complex relationships between climatic and agricultural variables, leading to reduced predictive accuracy. This study developed and evaluated machine learning–enhanced statistical models for predicting climate-related agricultural risks in Pakistan. Secondary data were obtained from the Pakistan Meteorological Department (PMD), the Food and Agriculture Organization (FAO), and relevant agricultural databases, incorporating key climatic variables such as temperature, rainfall, humidity, and soil moisture alongside crop yield data for major crops. A hybrid modeling framework integrating machine learning algorithms (Random Forest, Support Vector Machines, Artificial Neural Networks, and Gradient Boosting) with traditional statistical techniques was employed. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R²). The results revealed that machine learning models significantly outperformed traditional statistical approaches, while the hybrid model achieved the highest predictive accuracy (R² = 0.94). Findings further indicated that rainfall and temperature were the most influential predictors of agricultural risk, highlighting the critical role of climate variability in crop yield instability. The study concludes that machine learning–enhanced statistical models provide a robust and reliable framework for agricultural risk prediction and climate-resilient decision-making. The proposed approach can support early warning systems and improve agricultural planning and policy formulation in Pakistan.