Feature Selection Method Based on Dimensionality Reduction and Genetic Algorithm for Detecting Colon Cancer

Cancer is a deadly condition that claims lives all over the world. Finding a few genes pertinent to one type of cancer can lead to beneficial therapies. The difficulty with microarray databases is their high dimensionality; they include many characteristics relative to a small number of examples. Recent studies have tried to use various feature selection strategies to minimize this excessive dimensionality. An ensemble feature selection method based on dimensionality reduction and genetic algorithms is presented. A Genetic Algorithm-based Cancer Detection Framework (GA-CDF) is proposed to detect cancer using feature selection and classification methods. After preparing the data with a dimensionality reduction, the best subset of attributes is obtained by integrating the information from two independent databases using GA. On a database for colon cancer, GA-CDF is used with 5-fold cross-validation. The integrated feature selection method applies wrapper methods to obtain the minor optimum gene subset. After being confirmed on a different database, the genes subset achieved a classification efficiency of 98.2% for SVM and 97.5% for NB.