A COMPARATIVE ANALYSIS OF HYBRIDIZED GENETIC ALGORITHM IN PREDICTIVE MODELS OF BREAST CANCER TUMORS

Abstract

The death rate from breast cancer has been steadily declining, but recent advances in computer-aided methods for developing reliable early prediction models have been a boon. Research has shown that genetic algorithms are effective feature selection tools for data pre-processing and that random forest predictors outperform other machine learning regressors in terms of accuracy. Multiple studies have attempted to enhance breast cancer prediction models by creating hybridised genetic algorithm models for feature selection; however, it is unclear whether the order of hybridisation was considered, since this could affect the performance of the hybridised model. Consequently, this study suggests a number of effective prediction models that make use of a hybridised genetic algorithm; these models are built on various learning models, and they account for the hybridised models' feature selection algorithms' placement order. As a test bed, the Wisconsin Breast Cancer dataset was used, and the suggested hybridised models made use of filter, wrapper, and embedding feature selection techniques. In this study, we compared the individual learning models' results with those of the suggested hybridised models. The following models are part of this set: Fisher_Score, Mutual Information Gain, Chi-square test for correlation, coefficient, variance, genetic algorithm, ridge regression with L2 regularisation, tree-based approaches, and linear and Lasso regression with L1 regularisation. With an accuracy score of 99.12%, the suggested hybridised Genetic Algorithm with Fisher Score (GA + Fisher Score) model outperformed its competitors in the performance test. performing other proposed hybridized genetic algorithm models considered.