Feature selection to simplify BDI for efficient depression identification

Jiayue Cai, Z. Jane Wang, Silke Appel-Cresswell, Martin J. McKeown

Abstract: The Beck Depression Inventory (BDI), a self-report questionnaire consisting of 21 question items, has been the most extensively used for depression assessment. The problem of interest here is to identify a subset of questions in the BDI that are most predictive of depression and can reveal gender differences between depression profiles. We investigate feature selection techniques to select a subset of items in the BDI for depression identification and suggest a new simple decision-making rule that yields high classification accuracy. Distinct features in the BDI are identified for depression assessment in men and women. For males, the most useful features include mood, loss of libido, work inhibition, self-accusation, social withdrawal while indecisiveness, mood, work inhibition, distortion of body image, crying are most useful for depression identification in females.