A Novel Approach to Feature Selection Using Genetic Algorithm and Support Vector Machine Classification in lung Cancer Diagnosis

Lung cancer is one of the deadliest cancers, such that it causes more deaths compared to breast cancer, colon cancer and prostate cancer and it is mainly because it cannot be diagnosed at early stages due to shortage of symptoms, such that survival rate of patients for 5 years after surgery is only 14%; while diagnosing the disease at early stages increases this probability to 70%. Increasing growth of this disease, difficulty of its diagnosis from images and importance of diagnosis at early stages requires CAD methods with high accuracy. In order to realize this important, a novel algorithm is proposed in this paper which selects features online using genetic algorithm and statistical functions. Our purpose is to separate effective features among available features. In order to classify data, a series of data called feature is required for which disease features are used. In many datasets, some features do not affect decisions and they are additional. So selecting an appropriate subset of inputs can be effective in classification accuracy and its speed. For this purpose, genetic algorithm with an objective function based on data sparsity and statistical concepts. The proposed method is implemented and results indicate high accuracy of this algorithm in selecting effective features and increasing accuracy of the classifier compared to basic methods and other studies.

Keywords:

Feature selection , machine learning , statistical genetic algorithm , lung cancer diagnosis