Classical Machine Learning

Random Forest models performed efficiently with small parameter counts and low computation time, with AUC near or above 0.99. Classical comparisons used neural networks, SVM, and random forests across the three regulatory sample sets. Neural networks achieved the highest classical accuracy and F1 in some settings but required many more parameters and longer runtime. The study frames QNN performance against classical models mainly around parameter efficiency. The study repeats Random Forest training across random seeds to assess whether top predictors are consistent. SVM performance was unstable at higher dimensions in some settings. Random Forest models are used within the largest cluster to rank variables by predictive importance. The highest-ranked health attributes from Random Forest analysis are used in the longitudinal trajectory modelling phase.