Reliable Multi-Class Mental Health Prediction Using a WiSARD Discriminator Model on Imbalanced Data.

The WiSARD classifier achieved the highest overall performance among all tested models for multi-class mental disorder prediction.

• WiSARD achieved an overall accuracy of 98.27%, F-measure of 0.983, MCC of 0.982, and KS of 0.981.
• Baseline models compared included Multilayer Perceptron, Naïve Bayes, DTNB, IB1, and A1DE.
• Evaluation was conducted using a 10-fold stratified cross-validation design.
• All models were tested under the same evaluation conditions on the same dataset.

WiSARD demonstrated greater tolerance for misclassifications associated with minority disorder classes, addressing class imbalance in the dataset.

• Analysis of ROC-AUC, TPR, TNR, and error distributions showed WiSARD's relative robustness to minority class misclassification.
• The dataset contained multiple disorder classes including Major Depressive Disorder, Anxiety, PTSD, OCD, ADHD, and Bipolar Disorder.
• Class imbalance was identified as a key challenge in the multi-class classification problem.
• Performance metrics including precision, recall, F-measure, accuracy, MCC, MAE, and KS were used to assess imbalance tolerance.

An ablation study confirmed that RAM-based pattern recognition in WiSARD contributed to improved reliability and interpretability.

• The ablation study specifically verified the contribution of RAM-based pattern recognition to model performance.
• WiSARD is described as an interpretable, RAM-based classifier.
• The ablation study was used to isolate and confirm the contribution of specific model components.
• Interpretability was identified as a key advantage of WiSARD over other tested models.

The study used a dataset of 637 complete records with 29 symptom-based features representing multiple mental health disorders.

• The dataset was sourced from the publicly available Kaggle Mental Disorders Dataset.
• Records with missing values, incomplete diagnostic labels, or duplicated entries were excluded, resulting in 637 complete cases selected for analysis.
• Features were symptom-based and represented disorders such as Major Depressive Disorder, Anxiety, PTSD, OCD, ADHD, and Bipolar Disorder.
• No clinical identifiers were involved, and ethical clearance was not required.
• The study was retrospective, conducted in 2024, and geographically based in Pakistan as part of computational healthcare research.

The study findings are limited by reliance on a single non-clinical Kaggle dataset with self-reported observations and without formal psychiatric validation.

• The dataset consisted of self-reported observations rather than clinically validated diagnoses.
• No formal psychiatric validation was performed on the dataset.
• The authors state that 'results are limited to a single non-clinical Kaggle dataset with self-reported observations and without formal psychiatric validation.'
• The authors recommend that 'the findings should be interpreted as indicative rather than definitive.'