AMAZonn (A Multicollinearity-adjusted Adaptive LASSO for Zero-infated Count Regression) with Weight of Expectation Maximization Standard Error Adaptive LASSO (SEAL AL) for Zero Inflated Poisson Data

Abstract

Lasso as selecting predictor variables continue to experience development like an adaptive Lasso which gives weight value in its formula. In a modelling case, variables selecting technique is needed in order to get a stable model, however, multicollinear cases are often found in several cases which caused the models obtained are unstable since the values of variance became large. Besides, data counting on response variable, with the presence of excess zero, causes the linear model cannot be applied, hence, using generalized linier modelling with zero inflated poisson (ZIP) model can become the solution. Thus, in this research, ZIP model will be applied after selecting the variables through AMAZoon (A Multicollinearity-adjusted Adaptive LASSO for Zero-infated Count Regression) with Weight of Expectation Maximization Standard Error Adaptive LASSO (SEAL AL), and the comparison towards the results gained by ZIP model without prior variables selection will be done. The comparison was seen based on the value of the smallest Akaike Information Criterion (AIC). The data analysis revealed that there was multicollinear case in the data, and also ZIP model, after conducting the variables selection by using AMAZoon with Weigth SEAL AL, reached smaller AIC than ZIP model which having no variable selection. Therefore, ZIP model after the selection of variables by using AMAZoon with Weight of SEAL AL was better to be used when multicollinear happened.