Feasibility of Mental Health Triage Call Priority Prediction Using Machine Learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Speech Signal Processing

Background:

• Accurate call prioritization is crucial for mental health helpline efficiency and caller outcomes.
• Current triage relies on subjective clinical judgment, necessitating objective methods.
• Identifying high-distress callers promptly is vital due to significant mental illness morbidity and mortality.

Purpose of the Study:

• To investigate the use of machine learning (ML) for estimating mental health helpline call priority.
• To analyze voice properties, rather than spoken content, for call assessment.

Main Methods:

• Speech data from phone callers was processed using existing APIs.
• Features were extracted from raw audio and fed into deep learning neural networks.
• A classification model was developed to determine call priority levels from audio representations.

Main Results:

• A deep learning neural network architecture was developed for instant voice-based priority assessment.
• The model analyzed 459 mental health helpline call records.
• The final ML model achieved 92% balanced accuracy in identifying call priority.

Conclusions:

• The developed ML model provides an objective measure of call priority based on voice quality.
• Results indicate voice analysis can estimate caller demeanor (positive/negative).
• This priority level can be displayed via a web interface for real-time decision support.