Crime Forecasting Using a Hybrid BiGRU–Transformer Neural Architecture: A Deep Learning Approach on Austin Crime Data

Mallidi, Revanth Reddy (2025) Crime Forecasting Using a Hybrid BiGRU–Transformer Neural Architecture: A Deep Learning Approach on Austin Crime Data. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (978kB) | Preview

Abstract

Crime forecasting has become increasingly important as urban environments generate large volumes of dynamic, irregular, and non-linear crime data that traditional statistical models struggle to interpret. With cities trusting on data analytics to improve public safety there is a pressing need for models that can capture long-term dependencies, sudden spikes and complex behavioural trends in crime patterns. This study addresses this gap by developing a novel Hybrid BiGRU–Transformer (BiGRUMHSA) approach which has designed to enhance predictive accuracy and inference performance in time-series crime forecasting. Using a large-scale dataset from the Austin Police Department, the research applies systematic data preprocessing, feature engineering, 15-day rolling window sequence generation, and comparative experimentation across LSTM, MLP, GRU, BiGRU, and the proposed hybrid model. Results show that the BiGRU-MHSA significantly reduces prediction error MSE 7.289, MAE 2.131 and offers an optimal balance between accuracy and computational efficiency. The model advances the state of the art by combining bidirectional temporal learning with multi-head attention, enabling more context-aware and stable predictions than traditional deep learning baselines. In practice, this provides law-enforcement agencies with a more reliable and real-time tool for proactive resource allocation. However, challenges remain in addressing data noise, district-level variability, and the need for adaptive learning in evolving crime environments.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Razzaq, Muhammad Asif UNSPECIFIED
Uncontrolled Keywords:	Crime Forecasting; Time-Series Analysis; BiGRU–Transformer Model; Deep Learning; Predictive Analytics
Subjects:	H Social Sciences > HV Social pathology. Social and public welfare > Criminology > Crimes and Offences Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HT Communities. Classes. Races > Urban Sociology
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	08 Sep 2026 09:02
Last Modified:	08 Sep 2026 09:02
URI:	https://norma.ncirl.ie/id/eprint/9876

Actions (login required)

View Item