A Variable Ranking Method for Machine Learning Models with Correlated Features: In-Silico Validation and Application for Diabetes Prediction

When building a predictive model for predicting clinical outcome using machine learning techniques, the developers are often interested in ranking features according to their ability. A commonly used approach obtain robust variable is apply recursive feature elimination (RFE) on multiple resamplings of training set and then aggregate results Borda count method. However, presence highly correlated can deteriorate performance. In this work, we propose variant method based RFE that takes into account correlation between variables during procedure order improve performance features. The proposed algorithm tested simulated datasets which true importance known compared standard RFE-Borda According root mean square error estimated rank (i.e., simulated) importance, overcomes Finally, applied case study related development type 2 diabetes onset.