Hybrid Machine Learning Architecture for Predicting Multiple Mental Disorders Using Blood Biomarkers and Demographic Features

View at publisher Summarise with VIRAT
Authors: Samindi Edirisooriya, Nilanga Vithanage, Buddhika Abeyrathna
Cited by: 0
References: 5

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Early detection of mental health disorders remains a significant challenge, particularly due to their frequent misattribution to physical conditions during initial clinical encounters. This diagnostic delay can hinder timely intervention and negatively impact patient outcomes. Traditional diagnostic approaches rely heavily on self-reported symptoms and psychological assessments, which may be incomplete or biased especially when patients present with somatic complaints such as fatigue, headaches, or sleep disturbances. In such scenarios, physicians often prioritize physical evaluations, and mental health concerns may only be addressed after inconclusive test results and subsequent referral, prolonging the diagnostic process. To address this gap, we propose a machine learning-based clinical decision support system that leverages routinely available blood biomarkers including cortisol levels, white blood cell count, and hormone levels along with demographic data such as age and sex, to support earlier risk assessment for mental health conditions. Since these tests are commonly performed in early stages of medical evaluation, the system is positioned to provide actionable insights before psychiatric referral, enhancing the likelihood of timely mental health screening. Our approach utilizes a hybrid artificial intelligence architecture combining an autoencoder for unsupervised feature extraction with an XGBoost-based multi-output regression model, specifically designed to predict risk scores across five major mental health disorders: anxiety, depression, attention-deficit/hyperactivity disorder (ADHD), obsessive-compulsive disorder (OCD), and schizophrenia. The model was trained on a dataset of 2,500 clinical records drawn from both local hospital data and international biobank repositories. Under 10-fold cross-validation, it achieved an average Root Mean Squared Error (RMSE) of 0.1779 and R2of 0.8004, demonstrating strong predictive performance.The results suggest that blood-based biomarkers and demographic variables can serve as objective, accessible, and scalable features to support early mental health risk assessment in clinical practice settings clinical settings. This system is designed to augment not replace clinical judgment and can be integrated into existing healthcare workflows to facilitate faster and more accurate mental health screening.

Fields of Study & Subjects

Artificial Intelligence in Healthcare Machine Learning in Healthcare Emotion and Mood Recognition

Publication details

Journal / source
IEEE International Conference on Industrial and Information Systems (ICIIS)
Pages
472–477
Publication date
16 January 2026
Language
EN