AI-Driven Predictive Analytics for Early Detection of Mental Health Disorders in Criminal Justice Systems
International Journal of Medical Toxicology and Forensic Medicine,
Vol. 16 (2026),
1 January 2026
,
Page 1-11
https://doi.org/10.22037/ijmtfm.v16.53727
Abstract
Background: Mental health disorders among individuals involved in criminal justice systems require timely and reliable assessment. Automated predictive approaches may assist mental health screening by integrating behavioural, clinical, and textual information. This study proposes an explainable hybrid framework combining a Bidirectional Encoder Representations from Transformers (BERT)-based language model with a Quick Medical Reference (QMR)-style Bayesian network for mental health disorder prediction.
Methods: The proposed framework incorporates data preprocessing, class balancing, dimensionality reduction, and hybrid probabilistic reasoning. Synthetic Minority Over-sampling Technique (SMOTE) is applied to address class imbalance, while Principal Component Analysis (PCA) and Independent Component Analysis (ICA) are employed for dimensionality reduction. The BERT-based model generates initial disorder probabilities from symptom descriptions, which are subsequently refined using a QMR-style Bayesian network through iterative symptom-based evidence updating. The proposed framework is evaluated using the Survey of Inmates in Local Jails (SILJ) 2024–2025 dataset and compared with Logistic Regression, XGBoost, and Explainable Boosting Machine.
Results: The proposed BERT-QMR-style Bayesian framework achieved an accuracy of 95.08%, precision of 93.41%, recall of 91.62%, F1-score of 92.51%, and AUC-ROC of 98.34%. The model also achieved an RMSE of 0.074 with a computational time of 64.2 seconds. Compared with the performance before preprocessing, the proposed framework improved accuracy from 86.76% to 95.08% and AUC-ROC from 89.63% to 98.34%, demonstrating the potential contribution of the integrated preprocessing and hybrid prediction strategy.
Conclusion: The proposed hybrid framework demonstrates the potential of combining contextual language representation with interpretable Bayesian reasoning for mental health assessment in criminal justice settings. The integration of BERT, SMOTE, PCA, ICA, and QMR-style Bayesian inference provides a promising approach to predictive mental health screening. Further validation using independent datasets, clinically validated labels, and external evaluation is necessary before practical deployment.
- Machine Learning, Explainable AI, Quick Medical Reference (QMR), Mental Health Disorder (MHD), Criminal Justice System (CJS)
How to Cite
References
[1] Thomaidou MA, Berryessa CM. Mental illness as a sentencing determinant: A comparative case law analysis based on a machine learning approach. Crim Justice Behav. 2023;50(11):1-44. [DOI: 10.1177/00938548231170801]
[2] Prince JD, Wald C. Risk of criminal justice system involvement among people with co-occurring severe mental illness and substance use disorder. Int J Law Psychiatry. 2018;58:1-8. [DOI: 10.1016/j.ijlp.2018.02.002]
[3] Olawade DB, Ayoola FI, Ebo TO, Asaolu AJ, Egbon E, David-Olawade AC. Artificial intelligence in forensic mental health: A review of applications and implications. J Forensic Leg Med. 2025;113:102895. [DOI: 10.1016/j.jflm.2025.102895]
[4] Altaf A, El Amraoui A, Delmotte F, Lecoutre C. Applications of artificial intelligence in mental health: A systematic literature review. J Comput Inf Syst. 2023;63(5):1280-300. [DOI: 10.1080/08874417.2022.2143455]
[5] Razavi M, et al. Machine learning, deep learning, and data preprocessing techniques for detecting, predicting, and monitoring stress and stress-related mental disorders: Scoping review. JMIR Ment Health. 2024;11:e53714. [DOI: 10.2196/53714]
[6] Wagay FA, Jahiruddin. Classification of mental illnesses from Reddit posts using Sentence-BERT embeddings and neural networks. Procedia Comput Sci. 2025;258:1669-1676. [DOI: 10.1016/j.procs.2025.04.398]
[7] Nweke F, Azmee AA, Khan MAAH, Pei Y, Thomas D, Nandan M. A transformer-driven framework for multi-label behavioral health classification in police narratives. Appl Comput Intell. 2024;4(2):234-52. [DOI: 10.3934/aci.2024014]
[8] Lamba K, Rani S, Shabaz M. Explainable machine learning for mental health prediction from social media behavior: A nested cross-validation study with SHAP and LIME interpretability. Discov Ment Health. 2026;6(1):1-15. [DOI: 10.1007/s44192-026-00373-z]
[9] Dehbozorgi R, et al. The application of artificial intelligence in the field of mental health: A systematic review. BMC Psychiatry. 2025;25(1):1-18. [DOI: 10.1186/s12888-025-06483-2]
[10] Pushpa G, et al. An advanced AI framework for mental health diagnostics using Bidirectional Encoder Representations from Transformers with Gated Recurrent Units and Convolutional Neural Networks. Ing Syst Inf. 2025;30(1):213-20. [DOI: 10.18280/isi.300118]
[11] Pavez J, Allende H. A hybrid system based on Bayesian networks and deep learning for explainable mental health diagnosis. Appl Sci. 2024;14(18):8283. [DOI: 10.3390/app14188283]
[12] Kyrimi E, McLachlan S, Dube K, Neves MR, Fahmi A, Fenton N. A comprehensive scoping review of Bayesian networks in healthcare: Past, present and future. Artif Intell Med. 2021;117:102108. [DOI: 10.1016/j.artmed.2021.102108]
[13] Chatzistamatis S, et al. A fairness-aware and interpretable model for recidivism prediction. Algorithms. 2026;19(7):509. [DOI: 10.3390/a19070509]
[14] Goudarzi M, Mohammadi A, Amirpour A, Fazeli M, Nasiri MJ, Hashemi A, et al. Genetic diversity and biofilm formation analysis of Staphylococcus aureus causing urinary tract infections in Tehran, Iran. The Journal of Infection in Developing Countries. 2019;13(09):777-85. [DOI: 10.3855/jidc.11329]
[15] Farayola MM, Tal I, Saber T, Connolly R, Bendechache M. A fairness-focused approach to recidivism prediction: Implications for accuracy, trust, and equity. AI Soc. 2026;41(4):2783-801. [DOI: 10.1007/s00146-025-02452-1]
[16] Cavus M, Benli MN, Altuntas U, Sari M, Ayan H, Ugurluoglu YF. Transparent and bias-resilient AI framework for recidivism prediction using deep learning and clustering techniques in criminal justice. Appl Soft Comput. 2025;176:113160. [DOI: 10.1016/j.asoc.2025.113160]
[17] Bureau of Justice Statistics. Survey of Inmates in Local Jails (SILJ), 2024–2025. Washington, DC: U.S. Department of Justice; 2025. [Link]
- Abstract Viewed: 0 times
- 53727 Downloaded: 0 times