Breast Cancer Classification With Microarray Gene Expression Data Based on Improved Whale Optimization Algorithm

2. Related Work

In the microarray dataset, the high ratio between the huge dimension of the genes (features) and the few number of samples resulted in inaccurate and imbalanced cancer prediction. In common, most of the genes in the microarray data are uninformative and redundant. These types of the genes are to be identified with the machine learning technique called as feature subset selection. Though the feature selection techniques not only identify the significant genes, it also improves the classification accuracy. There are three approaches for feature selection: (i) Filter method, (ii) Wrapper method and (iii) Hybrid method. In the filter method, the feature importance is measured with properties of the dataset and order the features based on the relevance score (feature importance score). This method is simple and fast and not considering the correlation among the features. The wrapper method generally incorporates any predefined classification algorithm to search for and select the relevance features. This method considers the feature dependencies and computationally intensive and slower. The hybrid method is the combination of filter and wrapper methods. This method, first apply the filter technique to reduce the feature space then use the wrapper method for feature subset selection. Since the wrapper method is expensive, it is proved to be beneficial in finding feature subsets that suit a predetermined classifier (Alshamlan et al., 2015).