Using Machine Learning to Determine a Functional Classifier of Retaliation and Its Association With Aggression

Abstract

ObjectiveMethods to determine integrity of integrated neural systems engaged in functional processes have proven elusive. This study sought to determine the extent to which: (i) a machine learning retaliation classifier (retaliation vs. unfair offer) developed from a sample of typically developing (TD) adolescents could be applied to an independent sample of clinically concerning (CC) youth; and (ii) the classifier-determined functional integrity for retaliation was associated with antisocial behavior, proactive and reactive aggression. MethodBOLD response data were collected from 82 TD and 120 CC adolescents whilst they performed a retaliation task. The Support Vector Machine (SVM) algorithm was applied on the TD sample and tested on the CC sample (adolescents with externalizing and internalizing diagnoses). ResultsThe SVM was able to distinguish the offer from the retaliation phase after training in the TD sample (accuracy=92.48%, sensitivity=89.47% and specificity=93.18%) that was comparably successful in distinguishing function in the test sample. Increasing retaliation distance from the hyperplane was associated with decreasing conduct problems and proactive aggression. ConclusionsThe current study provides preliminary data of the importance of a retaliation endophenotype whose functional integrity is associated with reported levels of conduct problems and proactive aggression.