KMAN Publication Inc. (KMANPUB) Journal of Assessment and Research in Applied Counseling (JARAC) 8 Serial Number 28 2026 01 01 Machine-Learning Prediction of Therapeutic Alliance Quality from Client Readiness to Change and Affective Dysregulation Machine-Learning Prediction of Therapeutic Alliance Quality from Client Readiness to Change and Affective Dysregulation 1 10 10.61838/kman.jarac.5183 EN Journal Article 2025 10 04 <p><strong>Objective: </strong>The aim of this study is to utilize and compare advanced machine-learning algorithms to predict the early quality of the therapeutic alliance based on a multidimensional assessment of client readiness to change and specific facets of affective dysregulation.</p> <p><strong>Methods and Materials: </strong>A prospective observational design was utilized, involving n=452treatment-seeking adults initiating outpatient psychotherapy in Canada. Baseline predictor variables were collected prior to the first therapy session using the University of Rhode Island Change Assessment (URICA) and the Difficulties in Emotion Regulation Scale (DERS). The criterion variable, therapeutic alliance quality, was measured after the fourth session using the Working Alliance Inventory (WAI). Missing data (&lt;2.5%┤) were handled via multiple imputation. The dataset was standardized and split into an 80%training set (n=361) and a 20%testing set (n=91). Support Vector Regression (SVR), Multilayer Perceptron (MLP), and Random Forest (RF) models were trained and compared using randomized grid search with 10-fold cross-validation.</p> <p><strong>Findings: </strong>The sample had a mean age of M=34.6(SD=11.2) years. Preliminary analyses indicated that the WAI Total Score correlated positively with URICA (r=.46,p&lt;.01) and negatively with DERS (r=-.52,p&lt;.01). On the testing set, the Random Forest model demonstrated the highest predictive accuracy (R^2=.67, RMSE =7.15), outperforming the MLP (R^2=.56, RMSE =8.22) and SVR (R^2=.48, RMSE =8.95) algorithms. SHAP feature importance analysis of the RF model revealed that the strongest negative predictors of the alliance were the DERS Strategies (M.Abs.SHAP=2.84) and Goals (M.Abs.SHAP=2.45) subscales. The URICA Action stage (M.Abs.SHAP=2.10) and Precontemplation stage (M.Abs.SHAP=1.95) emerged as the most significant positive and negative motivational predictors, respectively.</p> <p><strong>Conclusion:</strong> Advanced machine-learning models can accurately forecast the early quality of the therapeutic alliance, computationally demonstrating that specific baseline deficits in emotion regulation and distinct motivational stages are fundamental determinants of the therapeutic bond.</p> https://journals.kmanpub.com/index.php/jarac/article/download/5183/9423