A Genetic Algorithm with Flexible Fitness Function for Feature Selection in Educational Data: Comparative Evaluation

Abstract

Educational Data Mining is an interdisciplinary field that helps understand educational phenomena through computational techniques. The databases of educational institutions are usually extensive, possessing many descriptive attributes that make the prediction process complex. In addition, the data can be sparse, redundant, irrelevant, and noisy, which can degrade the predictive quality of the models and affect computational performance. One way to simplify the problem is to identify the least important attributes and omit them from the modeling process. This can be performed by employing attribute selection techniques. This work evaluates different feature selection techniques applied to open educational data and paired alongside a genetic algorithm with a flexible fitness function. The methods and results described herein extend a previously published paper by: (i) describing a larger set of computational experiments; (ii) performing a hypothesis test over different classifiers; and (iii) presenting a more in-depth literature revision. The results obtained indicate an improvement in the classification process.