Comparing the Predictive and Classification Performances of Logistic Regression and Neural Networks: A Case Study on Timss 2011

Abstract

Investigating effective factors on students’ achievement has wide application area in educational studies. Specially, Trends in International Mathematics and Science Study (TIMSS) allows researchers to determine correlates of mathematics and science achievement for different countries. In this study, the predictive and classification performances of logistic regression and neural networks are compared to identify the impact levels of variables on students’ mathematics achievement in Turkey. Age, gender and scales created by TIMSS team for 8th grade students (students like learning, value learning, confident in math, engaged in math, bullied at school, home educational resources), are selected as predictive variables. Model fitting statistics show that two methods give similar results in prediction and classification. In addition to model results, students’ confidence is found as the most effective factor to improve mathematics achievement.