Comparison between best subset and lasso regression on consumer price index Malaysia

Abstract

The aim of this research is to determine the factors contributing to the prediction of the total Consumer Price Index (CPI) in Malaysia through model selection using two methods which are the best subset and LASSO regression. The outliers are identified using the leverage values and studentized deleted residuals while the multicollinearity variables will undergo progressive elimination identified through Variance Inflation Factor (VIF) values. Both methods were compared using the Mean Square Error of Prediction (MSE(P)) to find the best procedure to display the CPI data. The model with the smallest MSE(P) will be chosen as the best model. The result showed that the MSE(P) of the best model using both the best subset regression and LASSO regression is almost the same. Therefore, the model selection using LASSO regression will be chosen as the best approach due to the simple process of identifying the best model. The best LASSO model consists of nine major categories such as food and non-alcoholic beverages (X1), tobacco and alcoholic beverages (X2), footwear and clothing (X3), transport (X7), communication (X8), culture and recreation service (X9), education (X10), hotels and restaurants (X11), miscellaneous service and goods (X12).