Comparisons of possibility-and probability-based classification: an example of depression severity clustering

  • Authors:

  • Sen-Chi Yu;Yuan-Horng Lin

  • Affiliations:

  • Center of Teacher Education, Huafan University, Shiding, Taipei County, Taiwan;Department of Mathematics Education, National Taichung University, Taichung City, Taiwan

  • Venue:
  • FS'07 Proceedings of the 8th Conference on 8th WSEAS International Conference on Fuzzy Systems - Volume 8
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

The purpose of this study is to apply the Possibility-based (fuzzy c-means, FCM) and Probability-based Classification (Wald's method and k-means) to classification of severity of depression. The scoring of Beck Depression Inventory (BDI)-II of subjects were analyzed by clustering analysis while the diagnose of depression-severity by a psychiatrist was used as the criterion to evaluate classification accuracy. The empirical data of outpatient diagnosed as depression was given and the percentage of correct classification among FCM, Wald's method and k-means were compared. The analytical results show the Kendall's τ coefficient of FCM, Wald's method and k-means were .549, .316, and .395, respectively. That is, FCM exhibited a higher association between the original and classified membership than did Wald's and k-means methods. We concluded that FCM identified the data structure more accurately than the two crisp clustering methods. It is also suggested that considerable cost concerning prevention and cure of depression might be reduced via FCM.