Feature selection using multi-layer perceptron in HIV-1 protease cleavage data

Recently, several machine learning approaches have been applied to modeling of the specificity for HIV-1 protease cleavage domain. However, HIV-1 protease cleavage domain with high dimensionality and small number of samples could misguide classification modeling and its interpretation. Thus, a method to select a smaller number of relevant features is required. Appropriate feature selection could eliminate irrelevant and redundant features, and thus, improves prediction performance and provides faster and more cost-effective models. As a result, we can gain deeper insight about dataset. In this paper, we introduce a new feature selection method, called FS-MLP, that extracts relevant features using multi-layered perceptron learning. With the method, we could extract a set of effective features in a multi-variate and non-linear way. Our experimental results on three types of artificial datasets and HIV-1 protease cleavage dataset show that performance of the FS-MLP is higher than other methods.