Real-Time Crime Prediction in India Using Machine Learning

Abstract

This paper presents a machine learning-based system for predicting crime patterns across Indian cities using historical crime data from 2010-2024. By integrating Random Forest regression with geospatial analysis, the model achieves 92.7% accuracy (R² score) in forecasting crime rates per 100,000 population. The system processes 15+ crime categories across 19 cities, including Mumbai, Delhi, and Bengaluru, using features like population density, crime type, and temporal trends. A web-based dashboard provides interactive crime heatmaps, prediction visualizations, and comparative analytics for law enforcement agencies. Evaluation shows Mean Absolute Error (MAE) of 6.84 and Root Mean Squared Error (RMSE) of 9.4, outperforming baseline models like SVM (R²=0.52) and Decision Trees (R²=0.02). The work highlights AI\'s potential in proactive policing while addressing data bias and ethical challenges in predictive policing.

Introduction

India’s urban areas have seen rising crime rates between 2015 and 2022, notably a 28% increase in cybercrimes and 15% in violent crimes. Traditional policing struggles with resource allocation, creating a demand for data-driven solutions.

AI-Driven Crime Prediction System

• Objective: Predict city-wise crime rates using historical crime and socio-demographic data.

• Data: Includes NCRB records from 19 cities, covering 10 crime types (e.g., murder, cybercrime), population, and socio-economic indicators.

• Methodology:

  • Data preprocessing on 1,520 records from 2014–2021, including handling missing values, normalization, and encoding.

  • Machine learning models (Random Forest, SVM, Decision Tree) trained to predict crime rates.

  • Crime rate formula normalizes crime counts by population.

• System Architecture: Modular client-server setup with a responsive web interface for input and visualization, backed by Flask/Django APIs.

• Results:

  • Random Forest performed best with R² = 0.927, MAE = 6.84, RMSE = 9.40.

  • Model predictions closely matched actual crime rates, demonstrating strong accuracy and reliability.

  • Features like population, crime type, and year were most influential.

  • Interactive dashboard supports law enforcement and urban planning by identifying high-risk zones and predicting future crime trends.

Cyber Awareness Initiative

• Context: Increasing cybercrime risks (hacking, identity theft, fraud) pose threats in a rapidly digitizing India, where many users lack cybersecurity knowledge.

• Goal: Educate and empower users via a student-led platform that blends technology and awareness.

• Features:

  • Crime prediction via machine learning.

  • Face detection to identify potential threats.

  • Comprehensive guide to Indian cyber laws.

• Approach: Intuitive interface promotes active engagement, making cybersecurity knowledge accessible across age groups and professions.

• Mission: Foster safer digital behavior and prevention strategies through education and practical tools.

Key Insights from Data Analysis

• Crime data across Indian cities show skewed distributions, with a few cities exhibiting very high crime counts.

• Population and crime features vary significantly by region, necessitating careful data transformation for accurate modeling.

• Multi-dimensional data (temporal, spatial, demographic, and crime categories) enables nuanced understanding and prediction.

Conclusion

The Crime Rate Prediction research represents a significant step toward harnessing the power of machine learning to address pressing societal issues. By leveraging historical crime data and population statistics, the Random Forest-based predictive model demonstrated strong performance in forecasting city-wise crime rates with a high degree of accuracy (R² score: 0.927). This highlights the feasibility and reliability of AI-driven tools for proactive decision-making in public safety and urban governance. The model’s applicability extends across multiple domains—from aiding law enforcement in effective policing and resource allocation to supporting policymakers and urban planners in understanding crime dynamics. The inclusion of visual analytics further enhanced model interpretability and practical usability. However, the research also underscores certain limitations, including data incompleteness, underreporting, and the absence of key socio-economic indicators. These factors point toward essential areas for improvement and research. Looking ahead, the scope for future enhancement is vast. Integrating diverse real-time data sources (e.g., FIR APIs, social media feeds), incorporating deep learning models, ensuring explainable AI, and deploying the system through interactive dashboards can transform this prototype into a robust, real-world solution. Additionally, ethical considerations such as data privacy, fairness, and transparency will be central to responsible deployment. The current crime prediction model, while effective in forecasting trends for select urban centers, faces several notable limitations. Primarily, it suffers from data bias due to its reliance on reports from only 19 cities, excluding rural and semi-urban areas where crime often goes underreported. This geographic and demographic underrepresentation reduces the model’s applicability across India. Additionally, the model operates on static, historical data without real-time integration from police FIR systems or emergency call records, which limits its responsiveness to dynamic crime situations. Ethical concerns also emerge, as training the model on biased or incomplete data may inadvertently reinforce policing biases, especially in already marginalized communities, leading to unfair targeting and misallocation of resources. To overcome these challenges, future work will focus on expanding the dataset by incorporating socioeconomic indicators such as unemployment rates, literacy levels, and migration patterns, thereby offering a more holistic understanding of crime causation. From a technological standpoint, integrating Long Short-Term Memory (LSTM) networks can enhance the model\'s ability to capture temporal patterns and forecast short-term crime trends. Additionally, incorporating SHAP (SHapley Additive exPlanations) will improve model transparency by identifying how each feature influences predictions. This explainability is critical to fostering trust and ethical use, particularly when the system is employed in public safety and policy decisions. In conclusion, this work not only contributes a practical tool for crime prediction in India but also lays the foundation for a broader vision where data-driven insights can promote safer cities, informed governance, and socially responsible AI applications.

References

[1] National Crime Records Bureau. (2023). Crime in India Report. [2] Breiman, L. (2001). Random Forests. Machine Learning. [3] NIST. (2022). Ethical Guidelines for Predictive Policing.

Copyright

Copyright © 2025 Amrita Sarkar, Aman Kant Mishra, Udit Prakhar Singh, Shaquib Hassan, Goutam Singh Munda, Pritam Singh. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.