International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 11 | Nov 2025
p-ISSN: 2395-0072
www.irjet.net
A Systematic Review of Artificial Intelligence in Crime Prediction AISWARYA NS MSc Computer Science Student, St. Thomas (Autonomous) College, Thrissur 680001, Kerala, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The escalating complexity of criminal activities
Despite the promising potential of AI-driven crime prediction systems, their widespread adoption is contingent upon addressing their inherent limitations. A critical concern is the trustworthiness and interpretability of these complex models. Opaque "black-box" AI systems, while potentially highly accurate, can obscure the rationale behind their predictions, raising ethical questions regarding fairness, accountability, and potential biases, especially when applied to sensitive domains such as the criminal justice system. Moreover, challenges related to the availability of highquality, comprehensive, and ethically sourced crime data persist, hindering the development of universally applicable and robust predictive models for crime.
necessitates the use of advanced analytical tools for effective crime prevention and mitigation. This systematic review synthesizes recent advancements in artificial intelligence (AI), machine learning (ML), and deep learning (DL) techniques that have been applied to crime prediction. By analyzing over 150 scholarly articles, this study elucidates prominent methodologies, including Geographic Information System (GIS)-based, statistical, ML-based, and DL-based approaches, and examines the diverse datasets utilized, particularly emphasizing the increasing availability and utility of public crime records. Key findings reveal a predominant focus on spatio-temporal crime hotspot prediction and crime type classification, with Artificial Neural Networks and Random Forest emerging as frequently employed algorithms. Despite significant progress in predictive accuracy, substantial challenges persist, notably concerning data quality, accessibility, and the interpretability and explainability of AI models. Consequently, this review underscores the imperative for the development of transparent and trustworthy AI systems, which is deemed essential to foster public confidence and ensure their ethical deployment. Future research trajectories are proposed to address extant deficiencies through the integration of explainable AI (XAI) techniques, exploration of hitherto underutilized ML categories, and systematic development of more robust, ethically sound, and inherently interpretable crime prediction models.
This systematic review aims to provide a comprehensive synthesis of the current landscape of AI-based crimeprediction research. It meticulously examines the diverse methodologies employed, characteristics of the datasets utilized, and performance of various algorithms. Particular emphasis is placed on the emerging field of explainable AI (XAI) and its role in enhancing the transparency and trustworthiness of crime prediction models. By critically analyzing the existing literature, this study identifies key advancements and persistent challenges and outlines promising future research directions necessary for the responsible and effective integration of AI in crime prevention efforts.
II. LITERATURE SURVEY
Keywords: Crime prediction, Artificial Intelligence, Machine Learning, Deep Learning, Explainable AI, Spatio-temporal analysis, Data mining, Systematic Review.
I.
The domain of crime prediction has witnessed a significant evolution, transitioning from traditional statistical methods to advanced AI-driven approaches. Early methodologies primarily leveraged Geographic Information Systems (GIS) and hotspot analysis to visualize crime data and identify areas of concentrated criminal activity. While instrumental in strategic planning and resource allocation, these methods were often criticized for being more descriptive than truly predictive, primarily identifying areas where crime has occurred rather than forecasting where it will occur. Statistical techniques, such as AutoRegressive Integrated Moving Average (ARIMA) models, have historically provided valuable insights into temporal crime trends due to their inherent interpretability and ability to model time-series data. However, their capacity to account for the complex spatial dependencies and dynamic interactions inherent in crime data remains significantly limited, often treating locations as independent entities. To address these limitations, hybrid models, integrating the
INTRODUCTION
The pervasive nature of criminal activity poses significant societal and economic challenges worldwide. Traditional crime analysis methods, often reliant on manual investigation and reactive measures, frequently prove insufficient in adapting to evolving crime patterns and trends. The advent of artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), has revolutionized criminology by offering sophisticated tools capable of analyzing vast datasets to identify intricate patterns and anticipate future criminal occurrences. These technologies empower law enforcement agencies to transition from reactive responses to proactive crime prevention strategies, thereby optimizing resource allocation and enhancing public safety.
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 393