Evaluation and Comparison of Recidivism Modeling as a Survival Model
Ex-offenders may reoffend after completing their sentences (recidivism). In this study, survival analysis was used to identify factors that affect time to reoffend. The Accelerated Failure Time (AFT) model was compared with the Cox Proportional Hazard (Cox PH) model. The effects of certain covariates on the acceleration or deceleration of the time to recidivism were then explained by the AFT model. Conversely, the Cox PH model provided estimates of the relative risk (hazard ratios) between groups. The model was evaluated using Concordance Index (CI), Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC). CI was used to evaluate the model’s discrimination ability. The best-fitting model based on AIC and BIC was found to be the stratified Cox PH model after removal of several covariates. This was then used as the preferred model to estimate the relative failure risk between groups. Conversely, the AFT models gave higher CI values, indicating better predictive discrimination. The results suggest that the two models are of a different nature and should be selected depending on the analytical purpose and policy focus.
Authors
- Utriweni Mukhaiyar (ORCID: https://orcid.org/0000-0003-2353-7110)
- Andree Sulistio Chandra (ORCID: https://orcid.org/0009-0007-1504-7628)
- Putri Anugrah Ramadhani
- Rizky Uyuun Akfindarwan
- Citra Nauli Al Fariz
Institutions
- Bandung Institute of Technology (ID)
Publication Details
- Journal
- Journal of Mathematical and Fundamental Sciences
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5614/j.math.fund.sci.2026.58.1.5
- Primary Topic
- Statistical Methods and Inference
- Type
- article
- Field-Weighted Citation Impact
- 0.00