Hybrid random forest– artificial neural network model based forecasting of anthracnose in bottle gourd across different transplanting windows

• A hybrid Random Forest–Artificial Neural Network model was developed for anthracnose forecasting in bottle gourd. • Plant parts- and transplanting window-specific forecasts achieved high accuracy (R² > 0.88). • Minimum temperature and morning RH were key weather predictors with 2–3 week lags. • Forecast outputs were translated into threshold-based disease risk advisories. • The model supports climate-resilient, site-specific disease management strategies. Anthracnose, caused by Colletotrichum lagenarium , significantly impacts bottle gourd yield, especially under different climate conditions and transplanting times. This study used a hybrid Random Forest–Artificial Neural Network (RF-ANN) framework to forecast disease severity (Percent Disease Index, PDI) across four transplanting dates DOT (T 1 –T 4 ) and plant parts (leaves and fruits) during the 2023 and 2024 Kharif (monsoon) seasons. Weekly data on disease severity and weather (temperature, humidity, and rainfall) were collected and processed using z-score normalization and time-lag features. RF-based importance analysis identified minimum temperature, morning relative humidity, and rainfall as key predictors. These factors were used to train ANN models. Short-term forecasts (1–14 weeks) showed high model accuracy, with the best results at a 2-week lead time (R²: 0.88 for leaves and 0.89 for fruits; NRMSE: 13.21 and 12.64%, respectively). Long-term forecasts (1–5 months) remained stable, with RF-ANN outperforming standalone ANN and RF models (R² up to 0.86; NRMSE down to 14.72%). Predicted versus observed PDI values showed strong agreement across all DOTs and plant parts. Lag analysis indicated meteorological variables lagged by 2–3 weeks exerted the greatest influence on predictions. Risk thresholds were set to convert forecast results into actionable advisory categories: low (<10%), moderate (11–20%), high (21–30%), and very high (>30%) risk. This RF-ANN hybrid framework offers an effective machine learning tool for early anthracnose warning in bottle gourd. It supports climate-resilient agriculture by enabling timely, site-specific, and organ-specific disease risk advisories under variable climate and planting conditions.