Detection of Disease Through Clinical Data Using Light Intensity Based Firefly Algorithm

Abstract

In this study, we presented a hybrid gene selection model based on a novel multi-filter(MF) ensemble technique and light intensity difference-based Firefly Algorithm. A multifilter (ReliefF, Fisher Score, SU (Symmetrical Uncertainty), CFS (Correlation-based Feature Selection)) is applied to the original micro-array dataset to evaluate the best-ranked feature subsets individually. Ensemble multi-filer coupled with light intensity difference based firefly algorithm to achieve the optimized feature subset and at the last performance, evaluation is done using Support vector machine (SVM) by considering three different kernel parameters. From the experimental analysis, it is noteworthy to assure that the proposed methodology performs better than the existing approach. In this paper ensemble, a multi-filter method is used by implementing the concept of correlation-based redundancy and a Modified firefly algorithm (FA) called Light Intensity Difference Based Firefly Algorithm(LFA) is used as a wrapper to identify the optimal high-ranked feature subsets. So the selected subsets will be more robust for classification purposes. The correlation-based redundancy approach is used for the multi-filter case, where the top n genes of individuals are identified and a new dataset is prepared, which will be considered as input for the wrapper method. The main drawback behind considering this single filter may be trapped into the local optimum and may increase the complexity of the wrapper which may reduce the performance of the proposed model. Using multifilter (MF) not only considers all possible combinations of feature subsets by considering more than one filer but also provides a better opportunity to achieve better classification accuracy.