Feature selection and classification of protein subfamilies using Rough Sets

Machine learning methods are known to be inefficient when faced with many features that are unnecessary for rule discovery. In coping with this issue, many methods have been proposed for selecting important features. Among them is feature selection that selects a subset of discriminative features or attribute for model building due to its ability to avoid overfitting issue, improve model performance, provide faster and producing reliable model. This paper proposes a new method based on rough set algorithms, which is a rule-based data mining method to select the important features in bioinformatics datasets. Amino acid compositions are used as conditional features for the classification task. However, our results indicate that all amino acid composition features are equally important thus selecting the features are unnecessary. We do confirm the need of having a balance classes in classifying the protein function by demonstrating an increase of more than 15% in accuracy.