Modelling Crime Dynamics with a Spatio-Temporal CBAM Attention-Enhanced Deep Learning Model

Reeja Santhosh Kumar, Krishna (2025) Modelling Crime Dynamics with a Spatio-Temporal CBAM Attention-Enhanced Deep Learning Model. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (967kB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Crime prediction has increasingly relied on advanced spatiotemporal modelling due to the complex and dynamic nature of urban crime patterns. There are some traditional statistical and machine-learning approaches which struggle to capture long-range temporal dependencies, spatial interactions between regions, and contextual influences like weather which results in limited predictive reliability. This study addresses these challenges by developing a Spatio-Temporal CBAM Attention-Enhanced Deep Learning Model designed to improve the extraction of important spatial and temporal crime features. Using Chicago’s large-scale crime dataset combined with weather data, the study constructs and evaluates multiple baseline sequential models which includes LSTM, BiLSTM, BiGRU, and BiLSTM with Cross-Attention against the proposed dual-attention architecture.

Experimental findings show that the Spatio-Temporal CBAM model achieves the lowest MSE is 0.7036, RMSE is 0.8388, and MAE is 0.6685 outperforming all baselines by better focusing on informative spatiotemporal patterns while reducing noise. These results demonstrate both theoretical values, by advancing attention-based modelling for crime forecasting, and practical benefits, providing more reliable crime trend predictions for decision-makers. While the model improves accuracy and seasonal tracking, limitations persist in capturing extreme crime spikes and ensuring computational efficiency for real-time deployment. Future work should explore graph-based spatial learning, transformer architectures, and finer-granularity contextual data to further enhance predictive performance.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Sahni, Vikas UNSPECIFIED
Uncontrolled Keywords:	Crime prediction; Spatio-Temporal CBAM; Deep learning; Attention mechanisms; Forecasting models
Subjects:	H Social Sciences > HV Social pathology. Social and public welfare > Criminology > Crimes and Offences Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	08 Sep 2026 11:57
Last Modified:	08 Sep 2026 11:57
URI:	https://norma.ncirl.ie/id/eprint/9899

Actions (login required)

View Item