Mental health research increasingly relies on real-world clinical and survey data to understand patient preferences, treatment acceptability, and care outcomes. However, such data are frequently characterized by substantial class imbalance, reflecting unequal distributions across diagnostic categories, treatment modalities, and patient-reported preferences. This imbalance poses persistent methodological challenges for machine learning–based analysis, particularly when the objective is to model nuanced outcomes such as acceptability of psychotherapy, medication, or lifestyle-based interventions. The present study examines the role of contemporary data balancing techniques in improving the analytical reliability of machine learning models applied to mental health treatment preference data. Drawing on established theoretical frameworks for learning from imbalanced datasets and recent empirical studies in mental health informatics, this article synthesizes methodological insights with a focused application context relevant to low- and middle-income settings. The study outlines a structured analytical pipeline encompassing data preprocessing, sampling strategies, hybrid balancing approaches, and model evaluation under imbalance-sensitive metrics. Results indicate that appropriate balancing techniques are associated with improved minority-class representation and more stable predictive performance, particularly in settings involving patient acceptability and preference outcomes. The discussion situates these findings within broader debates on methodological rigor, ethical considerations, and interpretability in mental health machine learning research. The article contributes a comprehensive methodological perspective that supports more equitable and reliable computational analyses of mental health data, while acknowledging limitations related to generalizability, contextual variability, and evolving clinical practices.

Imbalanced datasets, mental health informatics, machine learning, treatment acceptability, data balancing techniques, patient preferences

1. Fernandez, A., Garcia, S., Galar, M., Patri, R.C., Krawczyk, B., and Herrera, F. Learning from Imbalanced Data Sets. Springer Nature Switzerland AG, ISBN 978-3-319-98074-4, 2018.

2. Batista, G.E.A.P.A., Prati, R.C., and Mornard, M.C. A study of the behavior of several methods for balancing machine learning data. SIGKDD Explorations, 6(1), 20–29, 2004.

3. Gentili, E., Franchini, G., Zese, R., Alberti, M., Domenicano, I., and Grassi, L. Machine learning from real data: A mental health registry case study. Computer Methods and Programs in Biomedicine, 5, 100132, 2024.

4. George, R.S., Mehrota, S., and Paulomi, M.S. Treatment acceptability and preference for psychotherapy and medication in patients with common mental disorders in an Indian tertiary care setting. Online Journal of Health and Allied Sciences, 20(4), 7, 2021.

5. Geeks for Geeks. Sample from a population using R. 2023.

6. Islam, M.T., and Mustafa, H.A. Multi layer hybrid (MLH) balancing techniques: A combined approach to remove data imbalance. Elsevier, 2023.

7. Ormeno, P., Marquez, G., and Taramasco, C. Evaluation of machine learning techniques for classifying and balancing data on an unbalanced mini-mental state examination test data collection applied in Chile. IEEE Access, 2024.

8. PTSD clinical guidelines. How do I choose between medication and therapy? American Psychological Association, 2017.

9. Richardson, K., Petukhova, R., Hughes, S., Pitt, J., Yucel, M., and Segrave, R. The acceptability of lifestyle medicine for the treatment of mental illness: Perspectives of people with and without lived experience of mental illness. BMC Public Health, 24, 171, 2024.

10. O’Callaghan, E., Belanger, H., Lucero, S., Boston, S., and Winsberg, M. Consumer expectations and attitudes about psychotherapy: Survey study. JMIR Formative Research, 7, e38696, 2023.