Fig. 2

Block diagram of the training and validation workflow. During model training statistical learning algorithms use cross-validation to fine tune their internal parameters in order to maximise MCC and estimate their predictive performance in new un-observed data. Multiple imputation by chained equation was used to impute missing values in the training datasets. A model has been trained in each imputed dataset and aggregated in the final model ensemble. During model validation the ensemble of trained models is used to predict un-observed instances. After outcome agnostic multiple imputation all trained models are being used to predict the outcome in all imputed validation sets. Then results are aggregated by majority voting