<p>Mustard (<i>Brassica juncea</i> L.) is a major oilseed crop in India, and its productivity is highly sensitive to climatic variability and disease incidence. Wide variations in sowing windows across locations and years create complex agro-environmental conditions that challenge accurate yield estimation. Reliable yield prediction under such variability is crucial for optimizing crop management and ensuring food and economic security. This study therefore aimed to evaluate and compare the performance of multiple statistical and machine learning approaches for predicting mustard yield in Uttarakhand using long-term (2006–2021) weather, yield and disease severity data. Weighted and unweighted weather indices were derived from weekly weather data during the crop growth period and used as predictors, individually and in combination with disease severity. Eight modelling approaches including ANN, ENET, ELM, LASSO, PCA-ANN, RF, Ridge, SVR, and stepwise regression were compared using RMSE, R², and Lin’s concordance correlation coefficient (CCC) to assess predictive performance. Results indicated that ANN and PCA-ANN outperformed other models, (R² = 0.81) when trained with unweighted indices. Incorporating weighting schemes enhanced model accuracy across all approaches, while adding disease severity further improved performance for linear models. Feature importance analysis identified temperature-, humidity-, and radiation-related indices as important predictors of mustard yield. Overall, our study shows that integrating weighted weather indices and disease severity into yield prediction models provides a robust framework for accurate yield forecasting, demonstrating their potential as reliable tools for data-driven yield forecasting in mustard-growing regions of Uttarakhand.</p>