KMAN Publication Inc. (KMANPUB) Journal of Assessment and Research in Applied Counseling (JARAC) 8 Serial Number 28 2026 01 01 Identifying Predictors of Treatment Dropout Using ML Analysis of Motivational Interviewing Processes and Attachment‑Related Avoidance Identifying Predictors of Treatment Dropout Using ML Analysis of Motivational Interviewing Processes and Attachment‑Related Avoidance 1 12 10.61838/kman.jarac.5194 EN Journal Article 2025 09 19 <p><strong>Objective: </strong>This study aimed to utilize advanced machine learning algorithms to evaluate whether baseline attachment-related avoidance and specific linguistic frequencies of client change talk and sustain talk during early Motivational Interviewing sessions can reliably predict subsequent outpatient psychotherapeutic treatment dropout.</p> <p><strong>Methods and Materials: </strong>The study employed a prospective, longitudinal observational design with a sample of N=485 adults from Mexico City. Dropout was defined as missing three consecutive sessions or mutually agreeing to terminate prematurely. Data collection included baseline demographics, the Experiences in Close Relationships-Revised questionnaire to assess attachment-related avoidance, and Motivational Interviewing Skill Code ratings derived from the audio-recorded and transcribed first two therapy sessions. Machine learning analysis incorporated k-nearest neighbors imputation, Synthetic Minority Over-sampling Technique (SMOTE) for class imbalance, and evaluated Random Forest, Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost) classifiers using stratified 10-fold cross-validation. SHapley Additive exPlanations (SHAP) values were utilized for model interpretability.</p> <p><strong>Findings: </strong>Results indicated that 157out of 485participants (32.4%) dropped out of treatment. Bivariate analyses revealed that dropouts had significantly higher attachment-related avoidance (M=4.25) compared to completers (M=3.15). During early sessions, the dropout cohort exhibited significantly more Sustain Talk (M=24.6 vs. M=15.2) and less Change Talk (M=19.8 vs. M=31.4), alongside receiving fewer complex reflections and lower empathy ratings from therapists. The XGBoost model demonstrated superior predictive performance (AUC=0.87), followed by Random Forest (AUC=0.84) and SVM (AUC=0.78). SHAP analyses identified attachment avoidance and patient sustain talk as the strongest overall predictors, highlighting a critical interaction effect where highly avoidant patients receiving fewer therapist complex reflections exhibited a compounding probability of dropout.</p> <p><strong>Conclusion:</strong> Machine learning algorithms can accurately predict treatment dropout from early clinical interactions, emphasizing that patient retention fundamentally depends on the complex interplay between severe attachment-related avoidance and the therapist’s technical proficiency in managing resistance.</p> https://journals.kmanpub.com/index.php/jarac/article/download/5194/9427