Hybrid Feature Selection Techniques Utilizing Soft Computing Methods for Cancer Data
Rabia Musheer Aziz, Amol Avinash Joshi, Kartik Kumar, Abdu Hamid Gaani
Abstract
Recently, various soft computing techniques have been used to extract k information from big data. A standardized format for evaluating the expression levels of thousands of genes is made available by DNA microarray technology. The cancers of several anatomical regions can be identified with the help of patterns developed by gene expressions in microarray technology . As the microarray data is too huge to process due to the curse of dimensionality problem. Therefore, in this paper, a hybrid machine learning framework using soft computing techniques for feature selection is designed and executed to eliminate unnecessary genes and identify important genes for the identification of cancer. In the first stage, the genes or the features are taken out with the aid of the higher-order Independent Component Analysis (ICA) technique. Then in the second level, Genetic Bee Colony (GBC) optimization techniques are utilized in this work for selecting the best genes or features before proceeding to the classification process. For the comparison purpose, three other optimization techniques are considered in this work, which are Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), and Genetic Algorithm (GA). After the selection of important genes, the 24 two most popular classifiers Naïve Bayes (NB) and Support Vector Machine (SVM)) are trained with selected genes and at last, find classification accuracy of test data. The experimental results with five benchmark microarray datasets of cancer, prove that GBC is a more efficient approach to improve the classification performance with ICA for both the classifiers.