Assessing the Effectiveness of a Genetic Programming Algorithm in Supervised Classification Tasks

Abstract

Evolutionary Algorithms (EAs) are population-based, stochastic search algorithms that emulate natural evolution. Over the years, EAs have been successfully applied to numerous classification problems. This paper proposes an enhanced evolutionary algorithm for synthesizing classifiers in supervised data scenarios. The algorithm is rooted in Genetic Programming (GP), a form of optimization inspired by biological evolution. GP evolves computer programs that perform specific tasks, represented as sets of genes manipulated by a Genetic Algorithm (GA). Our approach generates a Directed Acyclic Graph (DAG) for each training class to align each test class with one of them. Additionally, the approach incorporates a fitness evaluation function with dual objectives: the cumulative evaluation values of test samples and the count of prediction errors in classes. We benchmark our approach against established machine learning methods such as Support Vector Machine (SVM), K-Nearest Neighbor (K-NN), and Particle Swarm Optimization (PSO). The enhanced algorithm demonstrates competitive performance, achieving an accuracy of over 70% when evaluating evolving classifiers across twelve public UCI datasets.