Artificial intelligence in suicide prevention: Utilizing deep learning approach for early detection.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Psychiatry

Background:

• Student suicide rates in India are a growing concern, necessitating early identification for intervention.
• Proactive strategies are crucial for identifying vulnerable students, especially those preparing for competitive exams.
• Timely intervention can mitigate the risk of self-harm among students.

Purpose of the Study:

• To develop an artificial intelligence (AI) model using artificial neural networks (ANNs) for predicting suicidal tendencies in students.
• To create a technologically driven approach for identifying students at high risk of suicide.
• To facilitate early intervention for at-risk students, thereby reducing self-harm.

Main Methods:

• An AI model with an artificial neural network (ANN) architecture was designed for suicide risk prediction.
• A 33-feature input layer was curated from literature and expert insights, with weighted binary features.
• Hyperparameter optimization using the Optuna library and ridge regression were employed for model refinement and bias assessment. Model performance was validated against expert evaluations using statistical metrics and Cohen's Kappa coefficient.

Main Results:

• The AI model achieved exceptional predictive performance for suicide risk assessment in students.
• Key performance metrics include 98% accuracy, 100% precision, 97% recall, and a 98% F1 score.
• A Cohen's Kappa coefficient of 1.00 indicates substantial agreement between the model's classifications and expert evaluations.

Conclusions:

• The study presents a promising AI model utilizing ANNs for predicting suicide risk in stressed students.
• The model's high accuracy, precision, and recall, consistent with expert evaluations, support its potential for timely risk identification.
• The AI model's efficiency in assessing large student populations suggests significant clinical potential, warranting further refinement and real-world validation.