Enhancing Agricultural Decision-Making: A Hybrid Machine Learning Approach to Predicting Leaf Wetness Duration in Telangana

Abstract

Accurate estimation of Leaf Wetness Duration (LWD) is essential for effective agricultural management, particularly in optimizing spray programs and managing diseases influenced by microclimatic conditions. This study presents a novel approach by integrating Random Forest Regression (RFR) and Support Vector Regression (SVR) through a stacking ensemble technique to predict LWD. Using weather data from Telangana, India—comprising temperature, humidity, rainfall, and solar radiation—extensive preprocessing and feature engineering are applied before tuning the RFR and SVR models. A linear regression meta-regressor then combines the predictions from these models, improving LWD forecast accuracy. Performance is evaluated through cross- validation using metrics such as RMSE, MAE, and R². Preliminary results demonstrate that the stacked model outperforms traditional methods, providing more precise LWD predictions and supporting sustainable and productive farming practices in the region.