Analysis of the Certainty Factor Method for Depression Severity Classification in University Students

Abstract

Depression among university students requires accessible preliminary screening, yet the reliability of simplified rule-based models for representing multiple severity levels remains unclear. This study evaluates the performance of a Certainty Factor (CF) model for classifying depression severity using a publicly available secondary dataset containing 783 student records. After excluding four records with missing depression_severity labels, 779 records were analyzed. The model used three binary evidence variables depressiveness, suicidal ideation, and sleepiness and combined pre-specified Measure of Belief and Measure of Disbelief weights using the CF combination rule. The resulting CF scores were mapped to five predefined classes: None-minimal, Mild, Moderate, Moderately severe, and Severe. Model predictions were compared with questionnaire-derived depression-severity labels using a 5 × 5 multiclass confusion matrix and overall accuracy. The model correctly classified 315 of 779 records, producing an accuracy of 40.44% and an error rate of 59.56%. The evaluation also revealed that the model generated no Moderate predictions because none of the attainable CF scores fell within the Moderate interval. In addition, many Mild cases were classified as None-minimal because the three evidence variables did not represent the broader symptom profile captured by the reference questionnaire. These findings indicate that the evaluated CF configuration is insufficient for reliable five-class depression-severity classification. Future research should incorporate all PHQ-9 indicators, derive CF weights through a documented expert-elicitation process, justify missing-data handling, and validate the model on an independent dataset