Explainable Detection of Suicide Intent in Social Media Using Machine Learning and Transformer Models

Princess Chinemerem Iloh; Christian Arthur; The Anh Han; Raymond Chiong; Alessandro Di Stefano

Explainable Detection of Suicide Intent in Social Media Using Machine Learning and Transformer Models

Princess Chinemerem Iloh
, Christian Arthur
, The Anh Han
, Raymond Chiong
, Alessandro Di Stefano

Research output: Contribution to conference › Paper › peer-review

Abstract

Timely detection of suicidal ideation is vital for suicide prevention, as suicide continues to pose a significant global public health challenge. Social media often reflect early signs of suicidal intent, yet many predictive models lack transparency and interpretability. This study proposes an explainable AI framework combining traditional machine learning algorithms (Logistic Regression, Random Forest, SVM) and transformer-based models (BERT, DistilBERT) to detect suicidal intent in social media posts. SHAP is employed to enhance interpretability and provide insight into model decisions. Logistic Regression achieved 93% accuracy, while BERT attained 90%. The proposed framework offers transparent and actionable insights to support mental health professionals in early intervention efforts.

Original language	English
Publication status	Published - 5 Dec 2025
Event	18th International Conference on Multi-disciplinary Trends in Artificial Intelligence - MIWAI 2025: Multi-disciplinary Trends in Artificial Intelligence: 18th International Conference, MIWAI 2025, Ho Chi Minh City, Vietnam, December 3–5, 2025, - Ho Chi Minh , Ho Chi Minh , Viet Nam Duration: 3 Dec 2025 → 5 Dec 2025 https://miwai25.miwai.org/

Conference

Conference	18th International Conference on Multi-disciplinary Trends in Artificial Intelligence - MIWAI 2025
Abbreviated title	MIWAI 2025
Country/Territory	Viet Nam
City	Ho Chi Minh
Period	3/12/25 → 5/12/25
Internet address	https://miwai25.miwai.org/

Cite this

@conference{00c9f216be294ac395f6b2ea54f7a22c,

title = "Explainable Detection of Suicide Intent in Social Media Using Machine Learning and Transformer Models",

abstract = "Timely detection of suicidal ideation is vital for suicide prevention, as suicide continues to pose a significant global public health challenge. Social media often reflect early signs of suicidal intent, yet many predictive models lack transparency and interpretability. This study proposes an explainable AI framework combining traditional machine learning algorithms (Logistic Regression, Random Forest, SVM) and transformer-based models (BERT, DistilBERT) to detect suicidal intent in social media posts. SHAP is employed to enhance interpretability and provide insight into model decisions. Logistic Regression achieved 93\% accuracy, while BERT attained 90\%. The proposed framework offers transparent and actionable insights to support mental health professionals in early intervention efforts.",

author = "Iloh, \{Princess Chinemerem\} and Christian Arthur and Han, \{The Anh\} and Raymond Chiong and \{Di Stefano\}, Alessandro",

year = "2025",

month = dec,

day = "5",

language = "English",

note = "18th International Conference on Multi-disciplinary Trends in Artificial Intelligence - MIWAI 2025 : Multi-disciplinary Trends in Artificial Intelligence: 18th International Conference, MIWAI 2025, Ho Chi Minh City, Vietnam, December 3–5, 2025,, MIWAI 2025 ; Conference date: 03-12-2025 Through 05-12-2025",

url = "https://miwai25.miwai.org/",

}

Explainable Detection of Suicide Intent in Social Media Using Machine Learning and Transformer Models. / Iloh, Princess Chinemerem; Arthur, Christian; Han, The Anh et al.
2025. Paper presented at 18th International Conference on Multi-disciplinary Trends in Artificial Intelligence - MIWAI 2025, Ho Chi Minh , Viet Nam.

Research output: Contribution to conference › Paper › peer-review

TY - CONF

T1 - Explainable Detection of Suicide Intent in Social Media Using Machine Learning and Transformer Models

AU - Iloh, Princess Chinemerem

AU - Arthur, Christian

AU - Han, The Anh

AU - Chiong, Raymond

AU - Di Stefano, Alessandro

PY - 2025/12/5

Y1 - 2025/12/5

N2 - Timely detection of suicidal ideation is vital for suicide prevention, as suicide continues to pose a significant global public health challenge. Social media often reflect early signs of suicidal intent, yet many predictive models lack transparency and interpretability. This study proposes an explainable AI framework combining traditional machine learning algorithms (Logistic Regression, Random Forest, SVM) and transformer-based models (BERT, DistilBERT) to detect suicidal intent in social media posts. SHAP is employed to enhance interpretability and provide insight into model decisions. Logistic Regression achieved 93% accuracy, while BERT attained 90%. The proposed framework offers transparent and actionable insights to support mental health professionals in early intervention efforts.

AB - Timely detection of suicidal ideation is vital for suicide prevention, as suicide continues to pose a significant global public health challenge. Social media often reflect early signs of suicidal intent, yet many predictive models lack transparency and interpretability. This study proposes an explainable AI framework combining traditional machine learning algorithms (Logistic Regression, Random Forest, SVM) and transformer-based models (BERT, DistilBERT) to detect suicidal intent in social media posts. SHAP is employed to enhance interpretability and provide insight into model decisions. Logistic Regression achieved 93% accuracy, while BERT attained 90%. The proposed framework offers transparent and actionable insights to support mental health professionals in early intervention efforts.

UR - https://miwai25.miwai.org/#tentative_program

M3 - Paper

T2 - 18th International Conference on Multi-disciplinary Trends in Artificial Intelligence - MIWAI 2025

Y2 - 3 December 2025 through 5 December 2025

ER -

Explainable Detection of Suicide Intent in Social Media Using Machine Learning and Transformer Models

Abstract

Conference

Other files and links

Fingerprint

Cite this