A Study on Linkage-Based Clustering of Adult Depression Data

作者：Chattopadhyay Subhagata^*

来源：Journal of Medical Imaging and Health Informatics, 2013, 3(2): 227-236.

DOI：10.1166/jmihi.2013.1155

• Linkage
• Clustering
• Depression Data
• Symptoms
• Constructs
• Distance Measures
• Centroids

The paper aims to differentiate (or cluster) the major constructs (e.g., 'Emotional', 'Cognitive', 'Motivational', and 'Physical and vegetative') of depression-symptoms using linkage-based clustering, such as Hierarchical (having three subtypes: single, average and complete) and K-means (KM) techniques. Linkage-based techniques work by measuring distances among the clusters. Hence, three different distance measures, such as squared Euclidean (ED), City block (CB) and Cosine (COS) are used to investigate the best 'technique-distance' combination in obtaining the best clusters of the desired number. The algorithms are then tested on a sample of real-world depression cases (N = 302), each having fifteen quantified [0,1] symptoms and the corresponding depression probability (i.e., chance of depression). The symptoms are grouped under four constructs (mentioned above), each of which represents one cluster. The study observes that COS is the best distance measure for both the linkage-based techniques. 'KM-COS combination' is able to produce the best set of clusters. It also provides cluster center information, which could be useful to frame IF-THEN rules for automating the screening process. Finally, the clustering performance of KM-COS has been tested and the accuracy is found as 83.86%.

• 出版日期2013-6