Rat Swarm Optimizer based Transform for Performance Improvement of Machine Learning Classifiers in Diagnosis of Lung Cancer

Abstract

Usage of Machine Learning algorithms for assisting healthcare providers is increasing day by day. But the performance and robustness of the machine learning algorithms are the main concerns while implementing them for critical healthcare applications such as detection of cancer. This work concentrates on the performance improvement of supervised classifiers through the feature transform based on Rat Swarm Optimizer in diagnosing lung cancer using histopathological images. Rat Swarm Optimizer used for the transformation of features. These transformed features are more capable of providing better classification accuracy when compared to normal features. The dataset is downloaded from the publicly available website and three classes are present: normal, lung squamous cell carcinomas, and lung adenocarcinomas. In each class, 1000 histopathological images are considered. Four supervised classifiers namely Histogram-Gradient boosting classifier, Random forest classifier, K-Nearest Neighbor classifier, and Linear Discriminant Analysis classifiers are tested. The highest accuracy of 90.66% is offered by Histogram-Gradient boosting classifier and this is increased to 95.82% when Rat Swarm Optimizer is used as transform before classification.