Feature Selection Using CS - BPSO for Depression Detection Based on Profile Information

• Shaik Rasheeda Begum,
• Saad Yunus Sait,

Vol. 19, No. 3, March 31, 2025
10.3837/tiis.2025.03.003, Download Paper (Free):

• depression
• Machine Learning
• Feature selection
• Welch’s T-test
• cost-sensitive
• BPSO

Abstract

Depression is a common, cross-cultural mental health problem that, if unnoticed and untreated, can progress to a severe condition with serious consequences. Prevention and detection at an early stage are critical for overall health. We focus on detecting depression in college students using machine learning techniques, to identify individuals for a round of counseling, which is confirmatory. There is a possibility of normal individuals being identified as depressed (false positives) and depressed individuals being identified as normal (false negatives), each having a cost associated with them. Given that the costs of FP and FN are unknown and depend on the environment, we have proposed an algorithm that can provide different models for different relative costs of FP and FN. This work proposes a novel cost-sensitive feature selection algorithm using binary particle swarm optimization (CS-BPSO). The experiences and emotions of undergraduate students are collected through a private interview questionnaire form that allows participants to express their opinions freely while maintaining confidentiality. This is performed in conjunction with a Patient Health Questionnaire (PHQ-9) that permits labeling individuals as normal or depressed. The labeling is then validated manually by experts. The entire process has been performed under the guidance of psychiatrists. Results indicate that CS-BPSO consistently obtains the least cost compared to other feature selection algorithms, along with stability in classifier performance. Statistical tests also suggest that CS-BPSO performance is significantly better than other feature selection algorithms.

Statistics

Cite this article