In this research, the method of selecting explanatory variables in the linear regression model was presented in the presence of the two problems of high dimensions and outliers. The new selection method for variables was obtained by pooling or combining the Weighted Bootstrap with Probability method with the Robust LARS (RLARS) method called WBP-RLARS (Weighted Bootstrap probability - Robust Least Angle Regression method). It has been compared with some methods of selecting variables, namely B-LARS and B-RLARS Which depends on pooling or combining the pairs Bootstrap Approach in order to know the effect of employing some Bootstrap Approach (pairs Bootstrap and Weighted Bootstrap with probability) on the performance of the LARS method and the RLARS method in selecting variables in the HDRM model in the presence of outliers. The comparison was made using simulations, which were applied to different sample sizes (20, 26,70,90,104) and correlation coefficient (0.95) as well as data pollution ratios (0.05,0.10,0.15). Real data were analysed and the results of the analysis showed the preference of the WBP-RLARS method in the case of n < p and in the case of n>p the performance of the B-RLARS and WBP-RLARS method.
|