A Comparative Study on the Performance of Gene Expression Programming and Machine Learning Methods

Abstract

Usually, evolutionary algorithms are used to provide strong approximations to problems that are difficult to solve with other methods. Gene expression programming (GEP) is a type of evolutionary algorithm used in computer programming to generate computer programs or models. These computer programs are complex tree structures that, like a living organism, learn and adapt by modifying their sizes, shapes, and composition. In the present work, a comparison study was made among GEP and the standard prediction techniques to find the best predicting model on the BOSTON HOUSING dataset. Three approaches viz. GEP, ANN and polynomial regression were implemented on the dataset. The study showed how the three methods solve the problem of high bias and high variance and which one outperforms the other. The research work, however, gave a glimpse of the actual limitations and advantages of the methods on one another indicating the dependency of method on the type of data used. The results conclude the comparison of different methods on different performance metrics. The GEP model however reduced the problem of high bias and high variance by giving a slight difference between the train and test accuracy but was not able to outperform ANN and polynomial regression in terms of performance metrics.