Classification for Microarray Data Based on K-Means Clustering Combined with Modified Single-to-Noise-Ratio Based on Graph Energy

Abstract

This paper proposes a new classification for microarray data which utilizes K-means clustering combined with modified single-to-noise-ratio based on graph energy (SNRGE) method. This method is employed to select a small subset of characteristic features from DNA microarray data. Comparing with the single-to-noise-ratio (SNR) method proposed by Golub, it demonstrates that the SNRGES outperforms SNR method. SNRGE obtains significant improvement on the classification result via SNRGES in contrast with other SNR formulas, and the result shows that the use of SNRGE formula is critical in eliminating irrelevant features. As compared to other feature selection methods via five classifiers, the SNRGES yields better classification performance. On available training examples from four microarray databases, we indicate that SNRGES is capable of achieving better accuracies than previous studies, and is able to effectively remove redundant features and obtain efficient sets for sample classification purposes.