Skip to main content
WorldCist'18 - 6th World Conference on Information Systems and Technologies

Full Program »

A comparison of feature selection methods to optimize predictive models based on decision forest algorithms for academic data analysis

Nowadays, Feature Selection (FS) methods are essential (1) to create easy-to-explain predictive models in shorter periods of time, (2) to reduce overfitting and (3) avoid sparsity of data. The suitability of using these techniques is studied in this paper. Furthermore, a comparison of some widely extended techniques is performed to know which one is more appropriated to create predictive models using decision forest algorithms. For this comparison, experiments are conducted in which predictive models for each FS method are built to foresee if students will finish their degree after finishing their first year in college. A real dataset with students' data provided by the University of Almería is used to generate the predictive models. By comparing the accuracy of the built models, we can measure the effectiveness of each FS method, being the Chi-Square statistic the method that leads to better results in our experimental study.

Antonio Jesús Fernández García
University of Almería
Spain

Luis Iribarne
University of Almería
Spain

Antonio Corral
University of Almería
Spain

Javier Criado
University of Almería
Spain

 

Powered by OpenConf®
Copyright ©2002-2017 Zakon Group LLC