Variance Score and Pearson Similarity based Unsupervised Feature Selection (VPUFS) for Sample Clustering in Microarray Gene Expression Data

The analysis of high-dimensional gene profiles plays a vital part in cancer research, facilitating early detection, ameliorated prognostic, and the development of targeted therapeutic strategies. Grouping of similar data samples based on their gene profiles can be very beneficial for new cancer class identification. However, the high dimensionality and complexity of these gene profile data pose significant computational challenges, necessitating innovative approaches for feature selection and data reduction. This study proposes a novel framework called ‘Variance Score and Pearson Similarity based Unsupervised Feature Selection (VPUFS)’, a feature selection fashion to choose relevant features, for reducing high dimensionality of unlabeled datasets, thereby helping as a preprocessing step to cluster cancer samples effectively. The unsupervised dimensionality reduction framework leverages variance to calculate feature class relevance score, and similarity metrics to identify and retain informative features while eliminating the redundant ones. The proposed framework is rigorously evaluated across multiple gene profiles, such as, Leukemia, Prostate, Breast and Colon cancer samples. Comparative evaluations demonstrate that the proposed methodology outperforms existing feature selection techniques from unlabeled gene profiles, thus showing an enhanced stability and improved classification and clustering accuracies.