Predictive Analytics for City Crime Using Machine Learning

No Thumbnail Available
Almeida, Selina
Issue Date
MSc in Data Analytics
Dublin Business School
This study delves into the application of machine learning algorithms to predict urban crime, focusing on the dynamic and complex landscape of Los Angeles. Utilising a comprehensive dataset, the research explores the efficacy of various models like LSTM, GRU, SimpleRNN, Prophet, and XGBoost in predicting crime locations and types. The study aims to transition urban safety strategies from reactive to proactive measures by accurately forecasting crime patterns. The models were evaluated based on RMSE, MAE, and R² scores, with the data split into 80:20 and 70:30 ratios for training and testing. Results indicated that while LSTM, GRU, and XGBoost demonstrated high accuracy in spatial predictions, all models faced challenges in accurately predicting crime types, reflecting the multifaceted nature of criminal behaviour. The study highlights the potential of machine learning in enhancing urban safety but also notes the ethical and practical challenges inherent in predictive policing. It underscores the need for further research, especially in improving crime-type predictions and addressing the ethical implications of such predictive technologies. This research contributes significantly to the field of predictive urban crime analytics, offering insights and pathways for future innovations.