WITHDRAWN: MMFA-SVM: New bio-marker gene discovery algorithms for cancer gene expression

Abstract

Abstract Different Conventional machine learning algorithms like the gradient-based approach fail to achieve high accuracy with microarray datasets which are high dimensional and non-linear and fall with the local minima. To avoid these different metaheuristic approaches and nature-inspired swarm-based intelligent algorithms are adopted by various researchers for achieving global optima in the case of high dimensional datasets like the Microarray dataset. While the deal with microarray datasets as the size of the sample is very less as compare with feature size which is also directed towards the curse of dimensionality problem addition with various problems like redundancy, irrelevance, and noise. The extraction of significant biomarker genes is a vital task. This article proposes an innovative wrapper hybrid swarm intelligence approach called MMFA-SVM (Modified Mutated Firefly Algorithm – Support vector machine) for the identification of feature subsets to improve the accuracy of the proposed model. The proposed model works with two stages, in the first stage the enhancement of global convergence speed can be achieved using MMFA and in the second stage meta-search model is used with a well-known classifier SVM with LOOCV are adopted to calculate the accuracy of the biomarker feature subset. The search space exploration is done using MMFA with an enhancement approach of less glowing fireflies with more one. MMFA was used in this study with SVM to achieve optimal convergence time and a stochastic approach with the improvement of convergence time. The mutation-based firefly algorithm enhances the global search mobility of fireflies. We have compared the performance of the proposed one with FA, MMFA, MMFA-DT (decision tree), and MMFA-NB (Naive Bayes). From the simulation results, it confirms that MMFA-SVM performance is better as compared with normal FA, MMFA, MMFA-DT, and MMFA-NB with the microarray datasets. The efficiency of the proposed one performs better in comparison with its counterparts.