A novel approach of unsupervised feature selection using iterative shrinking and expansion algorithm

Abstract

An major constraint in the realm of feature selection is that users choose the ideal number of characteristics that must be picked. In this article, an effort is made to automate the process of determining a suitable value for the appropriate the quantity of characteristics that must be chosen for better recognition tasks. To estimate the ideal amount of features that should be maintained for properly describing the data, we use the dense subgraph discovery approach for this goal. Notably, the existing methods uses a similar kind of approach called the dense subgraph finding. But the earlier approach obtains a single dense subgraph, while the task of dense subgraphs finding obtains a number of dense subgraphs that are important for learning the internal structure of any network. Thus, dense subgraphs finding is likely to be adopted as a natural choice for realizing the complex relations among the nodes of massive real-life networks, such as biological networks, social networks.