Penerapan Algoritme Genetik Untuk Seleksi Peubah Regresi Logistik

Abstract

In a study, interaction factors are the potential to have important effects on the response variable. But research involving interaction factors often encounters two problems, namely the excessive number of variables and the difficulty of implementing the heredity principle. The alternative solution is to do variable selection using a metaheuristic optimization method, In this study, the logistic regression variable selection was done using a genetic algorithm. The genetic algorithm is modified so that every independent variable has a different probability to be included in the model. That probability is based on the absolute value of the correlation of the independent variable with the response variable. These modifications have a positive effect on the results of variable selection. To choose significant independent variables, 30 repetitions of the genetic algorithm can be performed using the objective function AIC. Of the 30 repetitions, if a variable appears in all formed models, then the variable is an independent variable that has a significant effect on the response variable. The application of this method to Myopia data can show significant variables well.