AbstractThree shrinkage regression and two machine‐learning approaches were evaluated to derive models for the prediction of epidemiological characteristics of white rust of mustard, using data from 112 epidemics in the field. Four epidemiological characteristics were considered: (a) crop age at first appearance of disease, (b) crop age at highest disease severity, (c) highest disease severity in a growing season and (d) area under disease progress curve (AUDPC), along with (e) crop yield to measure the effects of disease on crop performance. We developed models using weather indices to predict these variables using five different approaches: ANN, Elastic Net, LASSO, random forest and ridge regression. One model was developed for each sowing date corresponding to each dependent variable. Two hundred different models were developed. All models performed well at the calibration stage for most of the five variables at all sowing dates. However, at the validation stage, ANN‐derived models outperformed (R2val ~ 1.00, nRMSEV ~0.00 and MBEV ~0.00 in most cases) the three shrinkage regression‐derived models in predicting all five variables. Predictions by random forest‐ and LASSO‐derived models were acceptable for AUDPC and crop yield. Evaluation metrics (including R2val, nRMSEV and MBEV) suggested that ENET‐ and ridge‐derived models do not perform satisfactorily, whereas ANN‐derived models yielded reliable results and thus generate robust predictions. The present work constitutes a systematic effort to compare modelling methods for disease and yield prediction and illustrates the relevance of weather variables in predicting multiple epidemiological variables, and of multiple disease variables as predictors of actual crop yield.