Advancing Mental Health Risk Detection Through Transformer Ensembles

Yash Srivastava; Aditya Raj Singh; Tanisha Gupta; Suresh Kumar Poonia

Advancing Mental Health Risk Detection Through Transformer Ensembles

Author(s)	Mr. Yash Srivastava, Mr. Aditya Raj Singh, Ms. Tanisha Gupta, Dr. Suresh Kumar Poonia
Country	India
Abstract	This paper proposes a machine learning-based system that detects suicidal ideation automatically. This system will become the new solution to the challenges of large amounts of unstructured data and cross-domain generalization faced by efforts to monitor social media to identify suicide risk. As many existing Natural Language Processing (NLP) methods do not perform well when moving from well-formed text sources like Reddit into "noisy" environments such as Twitter, this framework will address this issue through a new model based on a modified Transformer architecture. The framework has two key components: a new form of Data Augmentation called a "Simulator," which will allow for data augmentation through the techniques of truncating and translating text and injecting emojis; and a model ensemble called a "Committee" using the three different pre-trained transformer architectures, RoBERTa, ALBERT, and DeBERTa, to promote maximum class discrimination and robust semantic interpretation of the data. Lastly, a loss function called "Heavy Hand," which applies a penalty of 10:1 for false negatives, will result in high recall. Thus, the architecture will be scalable, interpretable, and clinically safe for digital mental health monitoring.
Keywords	Suicidal Ideation Detection, Transformer Ensembles, Domain Adaptation, Data Augmentation, Cost-Sensitive Learning, RoBERTa, ALBERT, DeBERTa
Field	Computer > Artificial Intelligence / Simulation / Virtual Reality
Published In	Volume 16, Issue 4, October-December 2025
Published On	2025-12-08

View / Download PDF File

About IJSAT Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us	Message on WhatsApp	+91-9687-182-185	editor@ijsat.org

International Journal on Science and Technology

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Advancing Mental Health Risk Detection Through Transformer Ensembles

Share this