Publication

Abstract : The curse of dimensionality significantly hinders healthcare computational models due to the large size of the data. In this benchmarking study, we are using different feature selection and transformation methodologies to compare the computational efficiency and complexity of a Random Forest (RF) and Decision Tree (DT) classifier models for predicting heart disease risk. We investigate the performance of multicollinearity removal using Stratified cross-validation, Factor Analysis (FA), Principal-Component Analysis (PCA), and Linear-Discriminant Analysis (LDA) in terms of their ability to improve prediction accuracy and reduce computational time. The proposed RF model with appropriate feature engineering technique can produce a CHD risk prediction accuracy of 91% with only 7 features which is 8.33% higher than the preliminary model. Notably, we are also able to achieve a more than 33% reduction of time complexity in our best model as compared to the preliminary model. The statistical significance of our models’ results has been validated with a Z-test and Analysis of Variance (ANOVA). Our results provide insights into the trade-offs between prediction accuracy and computational efficiency when using these models for heart disease risk prediction. They have important implications for their practical use in clinical settings.