Distance Rank Score: Unsupervised filter method for feature selection on imbalanced dataset

This paper presents a new filter method for unsupervised feature selection. This method is particularly effective on imbalanced multi-class dataset, as in case of clusters of different anomaly types. Existing methods usually involve the variance of the features, which is not suitable when the different types of observations are not represented equally. Our method, based on Spearman's Rank Correlation between distances on the observations and on feature values, avoids this drawback. The performance of the method is measured on several clustering problems and is compared with existing filter methods suitable for unsupervised data.