Agriculture is a vital contributor to global food security but faces escalating threats from environmental fluctuations and pest incursions. Among the most prevalent and destructive pests, Jassid (Amrasca biguttula) and Thrips (Thrips tabaci) frequently afflict cotton, okra, and other major crops, resulting in substantial yield losses worldwide. This paper integrates five machine learning (ML) models to predict pest incidence based on key meteorological attributes, including temperature, relative humidity, wind speed, sunshine hours, and evaporation. Two ensemble strategies, soft voting and stacking, were evaluated to enhance predictive performance. Our findings indicate that a stacking ensemble yields superior results, achieving high multi-class AUC scores (0.985). To demystify the underlying mechanisms of the best-performing ensemble, this study employed SHapley Additive exPlanations (SHAP) to quantify the contributions of individual weather parameters. The SHAP analysis revealed that Standard Meteorological Week, evaporation, and relative humidity consistently exert the strongest influence on pest forecasts. These insights align with biological studies highlighting the role of seasonality and humid conditions in fostering Jassid and Thrips proliferation. Importantly, this explainable approach bolsters the practical utility of AI-based solutions for integrated pest management (IPM), enabling stakeholders—farmers, extension agents, and policymakers—to trust and effectively operationalize data-driven recommendations. Future research will focus on integrating real-time weather data and satellite imagery to further enhance prediction accuracy, as well as incorporating adaptive learning techniques to refine model performance under varying climatic conditions.