Online Program

HDSI: High-Dimensional Selection with Interactions Algorithm on Feature Selection and Testing (304189)

Keywords: LASSO, regression, interaction terms, bootstrapping, high-dimensional data, feature selection, HDSI

Classical statistical techniques face struggle of performing feature selection (FS) on high dimensional (HD) data along with interaction effects. Algorithms like random LASSO can handle HD data but neither considers interaction terms nor tests for significance of selected features. This study proposes High Dimensional Selection with Interactions (HDSI) algorithm to handle HD data, incorporate interaction terms, provide statistical inferences of selected features and leverage capability of existing classical statistical techniques. The method allows application of any statistical technique like LASSO and subset selection on multiple bootstrapped samples containing randomly selected features. Each sample incorporates interaction terms for the selected features. Each feature performance from different models is pooled and statistical significance is determined. The statistically significant features are selected. The final coefficients are estimated using appropriate statistical techniques. Simulated data and real studies are used to evaluate HDSI performance. In general, it outperforms common algorithms such as LASSO, subset selection, adaptive LASSO, random LASSO and group LASSO.