A stable gene selection method based on sample weighting

In microarray analysis, the selection of informative gene is an essential issue for tissue classification and successful treatment because of its ability to improve the accuracy and decrease computational complexity. However, the gene subsets selected by the same method often vary significantly with some variations of the samples in the same data set. Thus, the stability of the selected genes is an aspect as important as the predictive ability to quantify a feature selection algorithm. This work is an attempt to improve the stability of feature selection methods from the point of view of sample importance. We propose an efficient mean deviation-based sample weighting algorithm to improve the stability of feature selection methods by assigning a weight to each sample according to the mean deviation of its local value in giving samples. We perform a series of experiments with four frequently studied public data sets, and the experimental results validate that the proposed method improves the stability of common feature selection algorithms such as ReliefF without sacrificing the classification performance. Furthermore, more stable gene feature sets obtained by the proposed method than the state-of-the-art ensemble method and margin-based sample weighting algorithm.