Machine Learning Analysis of Case Clearance in Los Angeles Violent Crimes, 2020 to 2024

Shu Hong
MASDS, 2026
WU, YINGNIAN
This thesis explores whether the information contained in initial police reports canhelp predict whether a crime will be solved. I employed two different definitions of “case clearance”: a broad definition, under which any form of case closure is considered a clearance; and a strict definition, which counts a case as cleared only if an arrest is made. I ran three predictive models under each of these two definitions, using data on 20,553 violent crimes recorded by the LAPD between 2020 and 2024. Among three, Gradient Boosting method achieves the best calibration and lowest Brier score. A single behavioral indicator, showing whether the victim knew the suspect, accounted for roughly a third of total feature importance. Interestingly, victim descent contributed close to zero and observed group level disparities tracked the distribution of crime types rather than any learned ethnic pattern.
2026