A Random Forest Analysis Predicting Intergenerational Attachment Insecurity from Parental Reflective Functioning and Childhood Trauma Phenotypes

Abstract

Objective: The primary objective of this study was to utilize a Random Forest machine learning approach to predict the variance in intergenerational attachment insecurity using distinct childhood trauma phenotypes and multidimensional facets of parental reflective functioning.

Methods and Materials: A cross-sectional, correlational, and predictive design was employed with a community sample of Romanian primary caregivers of children aged to years. Data were collected via self-report instruments, including the Childhood Trauma Questionnaire (CTQ) to assess five trauma phenotypes, the Parental Reflective Functioning Questionnaire (PRFQ) to evaluate mentalizing dimensions, and the Experiences in Close Relationships-Revised (ECR-R) alongside the Child Attachment Questionnaire (CAQ) to index intergenerational attachment insecurity. During data preprocessing, missing values were resolved using K-nearest neighbors imputation, and independent variables were subjected to Z-score standardization. A Random Forest regression algorithm was trained on an data split and evaluated on the remaining testing set, with feature importance mathematically derived via the mean decrease in node impurity.

Findings: Descriptive statistics indicated that emotional neglect was the most frequently endorsed historical trauma phenotype ( , ), while participants generally demonstrated moderate to high levels of interest and curiosity in their children’s mental states ( , ). The standardized outcome variable of intergenerational attachment insecurity ( , ) showed significant positive bivariate correlations with emotional abuse ( , ) and emotional neglect ( , ). Feature importance extraction from the Random Forest model identified emotional neglect and emotional abuse as the most powerful distal trauma predictors. Among proximal psychological mechanisms, maladaptive pre-mentalizing modes emerged as the strongest positive predictor of insecurity ( , ), whereas adaptive reflective capacities—specifically interest and curiosity ( , ) and certainty about mental states ( , )—functioned as significant negative predictors.

Conclusion: Advanced machine learning successfully mapped the non-linear pathways of relational risk, revealing that historical emotional maltreatment and current pre-mentalizing deficits are the primary drivers of intergenerational attachment insecurity, while adaptive reflective functioning serves as a crucial protective buffer.