Stacking-Based Ensemble Framework and Feature Selection Technique for the Detection of Breast Cancer

Abstract

Breast cancer is the second most common cancer in women worldwide. The uncontrolled growth of breast cells is called breast cancer. The treatment of human breast cancer is a very critical process, and sometimes certain indicators may produce negative results. To avoid this misleading outcome situation, a reliable and accurate breast cancer diagnosis system must be available. The machine learning (ML) method is a modern and accurate technique that researchers have recently applied to predict and diagnose breast cancer. In this research article, we developed stack-based ensemble techniques and feature selection methods for the comprehensive performance of the algorithm and comparative analysis of breast cancer datasets with reduced attributes and all attributes. In this article, we first take the SVM, k nearest neighbors, Naive Bayes and perceptron as four ML algorithms as sub-models that have been trained and predicted from, and then combine them to make a new model called blending (stacking). Finally, logistic regression is used to predict the stacked model. It is significant that sub-models produce different results that are not correlated predictions. The stacking technique is best when all the sub-models are skillfully combined together. This article uses the five-feature selection technique because it affects the overall performance of the model. Unrelated or moderately related features may adversely affect the behavior of the model. After applying the feature selection method, now we have data set with reduced features as well as all features. We implemented logistic regression on a dataset with all features and a dataset with reduced features. Finally, we see that the dataset with reduced features has got improved accuracy.