Lists
Lists
Iruthayaraj, Shana Precilla, Homayounvala, Elaheh and Yates, Shaun S. (2026) Predicting latent victims using Explainable Boosting Machine (EBM): evidence from the crime survey for England and Wales 2022-2023. In: 14th International Conference on Frontiers of Intelligent Computing: Theory and Applications (FICTA-2026), 8-9 June 2026, London. (In Press)
Crime victimisation is a major social harm in England and Wales, but many serious incidents are not reported to the police. As a result, police-recorded crime data can underestimate the scale of harm, particularly among victims who experience serious offences but remain outside the formal reporting system. This study addresses this problem by developing an explainable machine learning pipeline to identify hidden high-harm victims using the Crime Survey for England and Wales 2022 to 2023 dataset, comprising 31,183 respondents and 8,826 victim form records. A novel binary target variable, Hidden High Harm, is constructed to identify individuals who experienced serious victimisation but did not report the incident to the police. The target is developed using LCA-based vulnerability profiling and principal component analysis to derive an underlying harm factor. An Explainable Boosting Machine is then used as the primary interpretable model and benchmarked against weighted logistic regression and XGBoost. The findings show that the Explainable Boosting Machine provides competitive predictive performance, achieving a PR-AUC of 0.55 and a ROC-AUC of 0.79. It also produces the lowest fairness disparities across ethnicity and disability protected attributes; these results suggest that interpretable machine learning can help identify hidden high-harm victimisation while supporting more transparent and fairness-aware decision-making.
The study makes five original contributions: a survey-weighted machine learning pipeline for CSEW data, an underlying harm factor, a non-reporting reason classifier, LCA-based vulnerability profiling, and a victim-facing algorithmic fairness audit within crime prediction research. Overall, the research argues that explainable machine learning can improve understanding of hidden victimisation and provide a more transparent basis for identifying serious harm that is not captured through police reporting alone.
![[thumbnail of 281_CRC.pdf]](https://repository.londonmet.ac.uk/style/images/fileicons/text.png "281_CRC.pdf")
Restricted to Repository staff only until 22 May 2027.
Download (919kB) | Request a copy
 |
View Item |