[Paper Review] Crime Prediction Using Machine Learning and Deep Learning: A Systematic Review and Future Directions

Varun Mandalapu, Lavanya Elluri+2 more|arXiv (Cornell University)|Mar 28, 2023

Crime Patterns and Interventions9 citations

TL;DR

This paper conducts a systematic review of over 150 studies on crime prediction using machine learning and deep learning, aggregating datasets, techniques, and trends, and proposing future research directions.

ABSTRACT

Predicting crime using machine learning and deep learning techniques has gained considerable attention from researchers in recent years, focusing on identifying patterns and trends in crime occurrences. This review paper examines over 150 articles to explore the various machine learning and deep learning algorithms applied to predict crime. The study provides access to the datasets used for crime prediction by researchers and analyzes prominent approaches applied in machine learning and deep learning algorithms to predict crime, offering insights into different trends and factors related to criminal activities. Additionally, the paper highlights potential gaps and future directions that can enhance the accuracy of crime prediction. Finally, the comprehensive overview of research discussed in this paper on crime prediction using machine learning and deep learning approaches serves as a valuable reference for researchers in this field. By gaining a deeper understanding of crime prediction techniques, law enforcement agencies can develop strategies to prevent and respond to criminal activities more effectively.

Motivation & Objective

• Summarize the landscape of machine learning and deep learning approaches applied to neighborhood crime prediction.
• Identify publicly available datasets used for crime prediction and their characteristics.
• Analyze trends in algorithms, feature types, and evaluation metrics across studies.
• Highlight gaps, challenges, and future directions to improve crime prediction accuracy and applicability.

Proposed method

• Employ a systematic literature review across IEEE, ACM, and ScienceDirect databases with targeted queries and wildcards.
• Apply automated filtering followed by manual screening to select relevant papers (approximately 157 in main text plus appendix).
• Perform pre- and post-analysis of literature, including a word cloud to identify key themes and a division of technique types by dataset source.
• Categorize studies by ML/DL techniques (classification, regression, clustering, etc.) and by data sources (crime, spatio-temporal, vision, social media, etc.).
• Summarize dataset resources and provide a table of datasets used (Table 1 reference in text).
• Discuss data-related challenges such as data quality, privacy, and interpretability of models.

Figure 1: Steps involved for typical crime detection

Experimental results

Research questions

• RQ1 What machine learning and deep learning techniques have been applied to crime prediction in recent literature?
• RQ2 What datasets and data sources are commonly used for neighborhood crime prediction, and what are their characteristics?
• RQ3 What are the prevailing trends in model types (classification, regression, clustering) and evaluation outcomes?
• RQ4 What gaps and future directions are identified to improve accuracy, interpretability, and real-time applicability of crime prediction models.

Key findings

• ML techniques dominate the literature (67%), followed by DL (21%), with smaller shares for ML+DL, DL+NLP, and ML+NLP.
• Classification is the primary task (63%), with regression (29%), clustering (6%), and mixed approaches (2%).
• A large portion of crime prediction studies appear in conferences (82%), with fewer in journals and other venues.
• A wide spectrum of datasets is used, including city-level crime data (e.g., NYC, Chicago, London) and global sources (surveillance video, social media, weather, etc.).
• Researchers report high accuracy in some contexts (e.g., up to 97% in certain Brazilian crime predictions; specific table entries note performance metrics across methods).
• The review highlights challenges: high-quality data availability, privacy/ethical concerns, and interpretability of complex models.

Figure 2: Research paper selection methodology