Machine learning–based hybrid framework for weather-driven forewarning of major groundnut pests

Abstract

Groundnut production in semi-arid regions is persistently constrained by major insect pests such as root grub, leaf miner, and earwig, whose population dynamics are strongly influenced by weather variability and crop seasonality. Accurate forecasting of pest incidence is therefore essential for timely and effective management interventions. This study presents a machine learning–based hybrid framework for weather-driven pest forewarning that integrates integer-valued time-series models with machine learning algorithms and exogenous weather variables to improve prediction accuracy. Using long-term light-trap data collected between 2004 and 2022 from the Regional Agricultural Research Station, Tirupati, Andhra Pradesh, the framework was implemented through hybrid model development by coupling INGARCHX with artificial neural networks, support vector regression, extreme learning machines, and random forest algorithms. The proposed hybrid framework consistently outperformed corresponding standalone models, demonstrating a superior ability to capture nonlinear, season-dependent pest–climate relationships. Within the framework, INGARCHX–SVR provided the most accurate forecasts for leaf miner, earwig, and kharif-season root grub populations, while INGARCHX–ANN performed best for rabi-season root grub. By combining count-based time-series modeling, machine learning, and weather covariates, the proposed framework strengthens early-warning capability, supports timely pest management decisions, and contributes to sustainable groundnut production in semi-arid agroecosystems.