Single and multiple outputs decision tree classification using bi-level discrete-continues genetic algorithm

Abstract

Data classification with decision tree models is an attractive method in data analysis and data mining. However, compared to other classification methods, the quality of prediction of these models is lower when classic heuristics and local optimization training methods are employed. To improve the performance of these models for single output and multiple outputs data sets, an optimal tree construction method based on the genetic algorithm is presented. The presented bi-level discrete-continues genetic algorithm method is able to select effective features as well as construct optimal tree. For this purpose, new operators of selection, crossover, and mutation are designed in terms of continuous and discrete variables. Comparison of the proposed method with other well-known classification methods for some test data sets and real world data shows that the performance of the decision tree models has been upgraded to the best of prediction methods level.