On Feature Selection Algorithms and Feature Selection Stability Measures:A Comparative Analysis

Data mining is indispensable for business organizations for extracting useful information from the huge volume of stored data which can be used in managerial decision making to survive in the competition. Due to the day-to-day advancements in information and communication technology, these data collected from e-commerce and e-governance are mostly high dimensional. Data mining prefers small datasets than high dimensional datasets. Feature selection is an important dimensionality reduction technique. The subsets selected in subsequent iterations by feature selection should be same or similar even in case of small perturbations of the dataset and is called as selection stability. It is recently becomes important topic of research community. The selection stability has been measured by various measures. This paper analyses the selection of the suitable search method and stability measure for the feature selection algorithms and also the influence of the characteristics of the dataset as the choice of the best approach is highly problem dependent.