Multi-filter Wrapper Enhanced Machine Learning Model for Cancer Diagnosis

The classification accuracy of the high dimensional dataset degrades due to the redundant and irrelevant features. Feature selection (FS) is used to reduce the dimensionality of the dataset by removing the noisy features. Each filter has its statistical approach. So the feature selected by a single filter may ignore the important one. We have presented a multifilter (MF) wrapper hybrid model. The advantage of using the MF method is to select the important feature by one filter which one may ignore by the other. Here, we have used an aggregator approach to combine the most efficacious features among the four individual filter methods (information gain (IG), chi-square (Chi-sq), minimum redundancy maximum relevance (mRMR), and relief). The accuracy assessment is carried out in a multiple filter wrapper (Jaya-SVM, GA-SVM, PSO-SVM, and FA-SVM). The evaluation and prediction of the subset of features are carried out with four classifiers with excellent performance, such as the support vector machine (SVM), Naive Bayes (NB), decision tree (DT), and linear discriminant analysis (LDA) were tested respectively. Four (breast cancer, leukemia, ovarian, and central nervous system (CNS)) cancer datasets are used to implement the model. The performance of the MF wrapper is excellent in comparison to a single filter. According to the findings of this study, the proposed hybrid approach is a more efficient and trustworthy feature selection technique for selecting highly discriminative features.