Analyzing Gang Reduction Strategies Through Dynamic Mode Decomposition

Abstract

We are interested in studying the effectiveness of the Gang Reduction and Youth Development (GRYD)’s prevention program by analyzing the change in estimated risk factor at different time steps. We view the change as a dynamical system of question responses per participant, and utilize the Dynamic Mode Decomposition algorithm to find the inherent temporal patterns. We are able to study youths' resiliency to change by their question responses, which may identify common themes that need to be addressed by the program. Furthermore, we develop a predictive model for the change in a youth’s attitudes and behaviors every six months of participation in the program. This model yields a similar predictive power as a shallow neural network but with faster run time and increased interpretability. We also observe transient growth in the system, indicating the program may be effective for youth who initially increase in risky behavior.

Report available on request.