Machine Learning Risk Prediction

Machine learning methods may improve risk prediction by identifying complex patterns in large datasets that traditional statistical approaches may not capture. Eight supervised learning classification algorithms were evaluated, comprising three linear models and five non-linear models. Hyperparameters were optimised with fivefold cross-validation, using area under the curve as the scoring metric. The dataset was split 80/20 into training and test sets, yielding approximately 1490 training patients with 203 events and 373 test patients with 51 events. Missing values were imputed using median values for continuous variables and mode values for categorical variables, with missingness below 5% for most variables.