TY - JOUR

T1 - Factors Associated With Momentary Acts of Aggression

T2 - An Investigation Using Machine Learning Approaches in Ecological Momentary Assessment Data

AU - Cheng, Y.

AU - Petrides, K.V.

AU - Ushakova, A.

AU - Ribeaud, D.

AU - Eisner, M.

AU - Murray, A.

PY - 2024/11/11

Y1 - 2024/11/11

N2 - Objective: Model the associations between aggressive behavior and potential precursors. Little research exists that can illuminate the most proximal factors to momentary aggression as they occur in daily life and against the background of an individual’s profile of relevant traits (e.g., their self-control levels). Method: This study used data from the combined longitudinal cohort and ecological momentary assessment (EMA) study, Decades-to-Minutes, with machine learning techniques to find the most important factors associated with “in-the-moment” aggressive behavior. Two types of models fitted by elastic net were examined: one with momentary data from the EMA component of the study and the other with both EMA and sociodemographic and trait data from the longitudinal survey component. Results: The best models fitted by elastic net achieved balanced accuracies of.76 and.79, while traditional methods achieved balanced accuracies of.63 and.64. Conclusions: Findings provide proof-of-concept evidence for the ability of elastic net to extract more important factors associated with aggression captured via short smartphone-based surveys and for the advantage of the elastic net method over stepwise regression for this purpose. The proposed models provide a step toward “in-the-moment” interventions to prevent aggressive behavior. Researchers are encouraged to apply the feature selection method used in this study for further research, such as exploring it in the context of smartphone applications for early prevention of aggressive behavior.

AB - Objective: Model the associations between aggressive behavior and potential precursors. Little research exists that can illuminate the most proximal factors to momentary aggression as they occur in daily life and against the background of an individual’s profile of relevant traits (e.g., their self-control levels). Method: This study used data from the combined longitudinal cohort and ecological momentary assessment (EMA) study, Decades-to-Minutes, with machine learning techniques to find the most important factors associated with “in-the-moment” aggressive behavior. Two types of models fitted by elastic net were examined: one with momentary data from the EMA component of the study and the other with both EMA and sociodemographic and trait data from the longitudinal survey component. Results: The best models fitted by elastic net achieved balanced accuracies of.76 and.79, while traditional methods achieved balanced accuracies of.63 and.64. Conclusions: Findings provide proof-of-concept evidence for the ability of elastic net to extract more important factors associated with aggression captured via short smartphone-based surveys and for the advantage of the elastic net method over stepwise regression for this purpose. The proposed models provide a step toward “in-the-moment” interventions to prevent aggressive behavior. Researchers are encouraged to apply the feature selection method used in this study for further research, such as exploring it in the context of smartphone applications for early prevention of aggressive behavior.

KW - aggressive behavior

KW - elastic net

KW - feature selection

KW - momentary ecological assessment

KW - supervised machine learning

U2 - 10.1037/vio0000556

DO - 10.1037/vio0000556

M3 - Journal article

JO - Psychology of Violence

JF - Psychology of Violence

SN - 2152-0828

ER -