Exploration of human activities using sensing data via deep embedded determination

Abstract

Clustering analysis is one of promising techniques of uncovering different types of human activities from a set of ubiquitous sensing data in an unsupervised manner. Previous work proposes deep clustering to learn feature representations that favor clustering tasks. However, these algorithms assume that the number of clusters is known a priori, which is often impractical in the real world. Determining the number of clusters from high dimensional data is challenging. On the other hand, the lack of the number of clusters make it difficult to extract low dimensional features appropriate for clustering. In this paper, we propose Deep Embedding Determination (DED), a method that can determine the number of clusters and extract appropriate features for the high dimensional real data. Our experimental evaluation on different datasets shows the effectiveness of DED, and the excellent performance of DED in exploring the human activities using sensing data.