A Predictive ML Model of Self‑Compassion, Shame‑Proneness, and Emotion Regulation Strategies in Counseling Outcomes

Abstract

Objective: The primary objective of this study was to construct and validate a predictive machine learning model to determine the relative mathematical feature importance and prognostic capacity of baseline self-compassion, shame-proneness, and emotion regulation strategies on outpatient clinical counseling outcomes. Methods and Materials: A prospective observational cohort design was utilized with n=452 Canadian adult outpatients attending a minimum of 6 psychotherapy sessions. Baseline psychological variables were assessed using the 26-item Self-Compassion Scale (SCS), the Test of Self-Conscious Affect (TOSCA) shame-proneness subscale, and the 10-item Emotion Regulation Questionnaire (ERQ). Ultimate treatment outcomes were quantified at termination using the 34-item Clinical Outcomes in Routine Evaluation-Outcome Measure (CORE-OM). Data preprocessing incorporated K-Nearest Neighbors imputation for missing values and Z-score normalization (z=(x-μ)/σ) for all psychometric features. Following dimensionality reduction, a Random Forest regressor was trained on an 80%training split (n=361) employing 10-fold cross-validation for hyperparameter tuning, and subsequently evaluated on a 20%testing split (n=91). Findings: The optimized Random Forest model (n_estimators=200, max_depth=10) demonstrated robust predictive generalization on the independent testing set, accounting for 61%of the variance in counseling outcomes (R^2=.61) with a Root Mean Square Error (RMSE) of 3.45and a Mean Absolute Percentage Error (MAPE) of 16.5%. The algorithmic extraction of Gini importance scores revealed a definitive predictive hierarchy: baseline self-compassion emerged as the paramount prognostic indicator (score =.385), followed sequentially by shame-proneness (.312), cognitive reappraisal (.195), and expressive suppression (.108). Conclusion: Baseline self-compassion and shame-proneness serve as the most critical transdiagnostic determinants of psychotherapeutic success, demonstrating that the application of machine learning algorithms to intake assessments can revolutionize personalized prognostic modeling and clinical treatment planning.