A-XGBoost: a resilient machine learning technique for predicting crimes against women across cultures on low cardinality crime data

A-XGBoost: a resilient machine learning technique for predicting crimes against women across cultures on low cardinality crime data

Violence against women is a global problem requiring innovative preventive measures. This research leverages Accelerated XGBoost (A-XGBoost), to predict crime against women in two culturally distinct countries: Finland and the United Arab Emirates (UAE). In culturally sensitive regions, public repor...

Full description

Saved in:
Bibliographic Details
Main Authors: Mathew Nicho, Ahmed Hamed, Tarek Gaber, Jamal Hamad Al Arimi
Format: Article
Language:English
Published: Taylor & Francis Group 2025-12-01
Series:Cogent Social Sciences
Subjects:
Violence against women
predicting crimes
machine learning
accelerated XGBoost
victim profiling
Artificial Intelligence
Online Access:https://www.tandfonline.com/doi/10.1080/23311886.2025.2527392
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Violence against women is a global problem requiring innovative preventive measures. This research leverages Accelerated XGBoost (A-XGBoost), to predict crime against women in two culturally distinct countries: Finland and the United Arab Emirates (UAE). In culturally sensitive regions, public reporting of violence against women is limited, leading to low-cardinality datasets. A-XGBoost addresses this by using Conditional GANs to generate realistic data samples while maintaining original feature distributions. This novel study proposes a culturally adaptive AI framework combining CGAN-based augmentation with optimized XGBoost , . Experimental results using A-XGBoost demonstrate 92.4% and 89.3% accuracy on UAE and Finland datasets respectively, with 6.7% and 3.1% improvements over traditional models. Under noisy conditions, A-XGBoost shows minimal performance degradation of 6.1% and 5.6%, outperforming similar studies. Its performance is further enhanced by automated hyperparameter tuning through grid search and cross-validation, ensuring optimal model configuration without excessive complexity. In Finland, the model highlighted intimate partner violence and trafficking, while in the UAE, public and sexual violence were dominant. This replicable framework tackles regional challenges, ensures strong predictive accuracy and upholding ethical standards by avoiding offender profiling. A-XGBoost maintains manageable time complexity, supporting real-world deployments. Hence predictive AI supports proactive interventions, resource optimisation, and evidence-based policymaking.
ISSN:2331-1886

Similar Items