Sparse group LASSO based uncertain feature selection

Abstract

Uncertain data management and mining is becoming a hot topic in recent years. However, little attention has been paid to uncertain feature selection so far. In this paper, we introduce the sparse group least absolution shrinkage and selection operator (LASSO) technique to construct a feature selection algorithm for uncertain data. Each uncertain feature is represented with a probability density function. We take each feature as a group of values. Through analysis of the current four sparse feature selection methods, LASSO, elastic net, group LASSO and sparse group LASSO, the sparse group LASSO is introduced to select feature selection from uncertain data. The proposed algorithm can select not only the features between groups, but also the sub-features in groups. As the trained weights of feature groups are sparse, the groups of features with weight zero are removed. Experiments on nine UCI datasets show that feature selection for uncertain data can reduce the number of features and sub-features at the same time. Moreover it can produce comparable accuracy with all features.