2.4. Logistic Regression Model Comparison

In comparison to ML models, the well-established LR method has been employed as the baseline approach to predict MACE in our dataset. The same raw features used for several models were used for the LR internal reference models. Remaining missing values were imputed: numerical features using mean or median training data and categorical variables using the mode of the feature as a global predictor and as a 1-year prediction. LR was configured with L1 and L2 regularization and maximum number of interactions 600.