Predicting working alliance in psychotherapy: A multi-modal machine learning approach.

Area of Science:

• Psychology
• Computer Science
• Artificial Intelligence

Background:

• Therapeutic alliance is crucial for treatment success, yet its consistent monitoring is challenging.
• Traditional self-report measures for working alliance lack widespread clinical implementation.
• Automated methods for alliance assessment could facilitate timely interventions.

Purpose of the Study:

• To develop an automated system for predicting the working alliance in therapy sessions.
• To utilize behavioral features from video-recorded sessions for alliance prediction.
• To investigate the efficacy of multimodal machine learning models for this task.

Main Methods:

• A dataset of 252 therapy sessions (in-person and teletherapy) with patient-rated working alliance was analyzed.
• Text, audio, and video features were extracted from session recordings.
• Machine learning regression models, including Gradient Boosting, were trained using fused multimodal features.

Main Results:

• The Gradient Boosting model achieved the highest accuracy using audio, text, and video features from the patient (ICC=0.66, Pearson r=0.70, MAE=0.33).
• Multimodal feature fusion significantly enhanced prediction performance.
• Automated alliance prediction demonstrated high accuracy in this study.

Conclusions:

• Automated alliance prediction from video-recorded therapy sessions is a feasible and accurate approach.
• Data-driven multimodal feature extraction and selection yield powerful predictive models.
• This technology has the potential to improve clinical practice by enabling early detection of alliance issues.