Machine learning prediction of conduct problems in children using the longitudinal ABCD study.

Area of Science:

• Child psychology
• Machine learning applications
• Developmental neuroscience

Background:

• Children with conduct problems face significant risks for negative psychosocial, educational, and behavioral outcomes.
• Early identification of at-risk children is crucial for timely intervention and improved long-term results.
• There is a need for effective screening tools to identify children who would benefit from early intervention for conduct problems.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting conduct problems in children.
• To identify key predictors of conduct problems using longitudinal data.
• To assess the accuracy and feasibility of using self-report data for early risk identification.

Main Methods:

• Utilized data from the longitudinal Adolescent Brain Cognitive Development (ABCD) Study.
• Employed machine learning classifiers including logistic regression, Naïve Bayes, support vector machine, and random forest.
• Predicted conduct disorder or oppositional defiant disorder over 1, 2, and 3 years in children aged 9-10 at baseline.

Main Results:

• The random forest classifier achieved high accuracy (AUC=0.98 at 1 year, 0.97 at 2 and 3 years).
• A simplified random forest model with 10 features demonstrated comparable predictive performance (AUC=0.97 at 1 year, 0.96 at 2 years, 0.97 at 3 years).
• Models achieved over 90% accuracy in predicting children at risk for conduct problems.

Conclusions:

• Machine learning models effectively identify predictors of conduct problems in children over a 3-year period.
• A limited set of self-report features can predict persistent conduct problems with high sensitivity and specificity.
• Parent and child self-report data, combined with machine learning, offers a promising approach for identifying at-risk children.