Machine Learning Techniques for Classification and Prediction of Pink Bollworm (Pectinophora gossypiella) Incidence Using Weather Variables at Guntur in Andhra Pradesh, India

Abstract

Pink bollworm, Pectinophora gossypiella, is an important insect pest of cotton that affects yield and fibre quality. The present study evaluated machine learning techniques for classifying and predicting pink bollworm incidence using weather variables at Guntur, Andhra Pradesh. Secondary data on pink bollworm incidence and weekly meteorological variables were collected from the Regional Agricultural Research Station, Lam, Guntur, for the period 2008–2024. The dataset included 393 observations and five weather variables: maximum temperature, minimum temperature, rainfall, morning relative humidity, and evening relative humidity. Pink bollworm incidence was categorised into low, medium, and high classes based on observed incidence levels. Descriptive statistics, Pearson correlation analysis, and four classification models, namely multinomial logistic regression, decision tree, random forest, and artificial neural network, were used for analysis. The descriptive results showed high variability in pink bollworm incidence, with a mean value of 9.81 and a coefficient of variation of 197.27%. Correlation analysis indicated that morning relative humidity had a significant positive association with pink bollworm incidence, whereas maximum temperature, minimum temperature, and rainfall showed significant negative associations. Among the evaluated models, Random Forest performed best, with training accuracy of 90.56%, testing accuracy of 85.78%, overall accuracy of 86.35%, Kappa value of 0.807, and AUC of 0.814. Decision Tree, Artificial Neural Network, and Multinomial Logistic Regression showed comparatively lower predictive performance. The results indicate that Random Forest can be a useful model for weather-based classification of pink bollworm incidence under the conditions of the present study.