Hybrid TABU search with SDS based feature selection for lung cancer prediction

Abstract

Cancer falls under a group of diseases where abnormal growths of the cells are observed. Generally, lung cancer does not result in any type of obvious symptoms in its early stages. Among the people diagnosed with lung cancer, about 40% are found to be in an advanced stage. Thus, the motivation of the work is to present an automatic screening of lung images for early diagnosis. For this, Machine Learning (ML) methods are popularly employed as a tool among medical researchers for classifying their medical images. To improve the performance of Lung cancer detection with ML techniques, feature selection is employed. As the feature selection is a Nondeterministic Polynomial (NP) hard problem, metaheuristic algorithms are widely used for finding the optimal feature set. The Tabu Search (TS) is semi-deterministic and also tends to act as a method of local, as well as global search. The techniques are capable of discovering and further identifying the relationships and patterns among them obtained from complex datasets and are also capable of effective prediction. In this work, a new hybrid TS with Stochastic Diffusion Search (SDS) based feature selection that was employed using the Naïve Bayes, Decision tree and Neural Network (NN) classifiers to improve classification. The results demonstrate the effectiveness of the proposed TABU-SDS- NN which achieves an accuracy of 94.07%.