Machine Learning Techniques for Classification and Prediction of Rice Leaf Folder (Cnaphalocrocis medinalis) Incidence Using Weather Variables

Rice (Oryza sativa) is a staple food crop in India and across Asia, but its production is threatened by the rice leaf folder, Cnaphalocrocis medinalis (Guenee) (Lepidoptera: Pyralidae), a major pest capable of causing yield losses of 63–80% under severe infestation. Climate change further exacerbates pest outbreaks, posing challenges to food security. This study applied machine learning classification algorithms including Multinomial Logistic Regression, C5.0 Decision Tree, Random Forest, and XGBoost to analyze rice leaf folder incidence based on five weather parameters. Secondary data were collected from the Agricultural Research Station, Bapatla, spanning 390 Standard Meteorological Weeks (SMWs) from January 2011 to December 2024 including weekly leaf folder catches and climatological averages: maximum temperature (T_max), minimum temperature (T_min), morning relative humidity (RHM), evening relative humidity (RHE), and total rainfall (RF). Multinomial Logistic Regression achieved the lowest accuracy (73.08%), while C5.0 Decision Tree accuracy improved to 83.08% with T_max as the most influential factor. XGBoost identified T_min as the key contributor achieving 93.59% accuracy. Random Forest outperformed all models yielding the highest accuracy (96.67%), a kappa value of 0.9155 and superior sensitivity, specificity, precision and F₁ scores. These results demonstrate that Random Forest is the most effective algorithm for predicting rice leaf folder incidence using weather data which offers a reliable tool for pest management and mitigation under changing climatic conditions.