Compactness score: a fast filter method for unsupervised feature selection

The rapid development of big data era incurs the generation huge amount day by in various fields. Due to large-scale and high-dimensional characteristics these data, it is often difficult achieve better decision-making practical applications. Therefore, an efficient analytical method urgently necessary. For feature engineering, selection seems be important research topic which anticipated select “excellent” features from candidate ones. implementation can not only purpose dimensionality reduction, but also improve computational efficiency result performance model. In many classification tasks, researchers found that seem usually close each other if they are same class; thus, local compactness great importance for evaluation a feature. Based on this discovery, we propose fast unsupervised algorithm, named Compactness Score (CSUFS), desired features. To prove superiority proposed several public sets considered with extensive experiments being performed. presented applying subsets selected through different algorithms clustering task. tasks indicated two well-known metrics, while reflected corresponding running time. As demonstrated, our algorithm more accurate compared existing