Artificial neural network coding of the child attachment interview using linguistic data.

Area of Science:

• Developmental Psychology
• Clinical Psychology
• Computational Psychiatry

Background:

• Adolescent attachment security is crucial, with insecurity linked to psychopathology.
• The Child Attachment Interview (CAI) is a valuable tool but is labor-intensive to code, limiting its accessibility.
• Developing automated methods for CAI coding could enhance its utility in research and clinical settings.

Purpose of the Study:

• To develop and evaluate computerized neural network models for predicting Child Attachment Interview (CAI) classifications.
• To assess if these models can meet the CAI's benchmark qualification of 80% accuracy on reliability training cases.
• To explore the feasibility of automated CAI coding for improved efficiency in adolescent attachment assessment.

Main Methods:

• Utilized 490 inpatient adolescent interviews for training and testing neural network models.
• Evaluated model performance on the CAI's 30 standard reliability cases as an independent holdout sample.
• Compared model accuracy and classification rates against human coders and the 80% benchmark.

Main Results:

• Neural network models demonstrated moderate accuracy and high correct classification rates compared to human coders on the general sample.
• Model performance was lower when applied to the CAI's standard reliability training cases.
• The 80% benchmark for reliability training cases was not consistently met by the developed models.

Conclusions:

• Automated coding of the Child Attachment Interview (CAI) using neural networks shows potential for improving assessment efficiency.
• Further development is needed to enhance model performance, particularly on standardized reliability cases.
• This approach could significantly increase the accessibility of attachment assessments in adolescent psychopathology research.