Suleiman Ibrahim / Afr.J.Bio.Sc.

6(Si2) (2024) 117-125 ISSN: 2663-2187

https://doi.org/ 10.33472/AFJBS.6.SI2.2024.117-125

Predicting Violent Crime Hot-Spots Utilizing Machine Learning

Suleiman Ibrahim1, Dipak Shankar Raskar2, Amit Jaykumar Chinchawade3, Sarang
Maruti Patil4, Madhav Jagannath Salunkhe5, Priyanka Sarang Patil6, Snehal Dipak
Raskar7, Paresh Jain8, Mukesh Bhardwaj8
1
Research Scholar, Computer Science and IT, Suresh Gyan Vihar University, India
2
Assisatant Professor, Department of CSE, Amity University, Mumbai
3
Assistant Professor, Department of AI and DS, SITCOE, Yadrav, Ichalkaranji. Maharashtra
4
Associate Professor, Department of Computer Engineering, SKN Sinhgad Institute of
Technology & Science
5
Associate Developer, Department of AI, Jeevan Technologies Chennai
6
Assistant Professor, Department of Information Technology, SKN Sinhgad Institute of
Technology & Science
7
Assistant Professor, Department of AIML, ViMEET, Khalapur.
8
Assistant Professor, ECE, Suresh Gyan Vihar University, India

Email: [email protected], [email protected],

Abstract: This research endeavours to develop an advanced model

Article History for predicting violent crime hot-spots by harnessing the power of
Volume 6,Issue Si2, 2024 machine learning techniques. The primary objective is to enhance
Received:23 Feb 2024 public safety measures by identifying areas prone to violent criminal
activities. The proposed model integrates a comprehensive set of
Accepted : 15 Mar 2024
spatial, temporal, and contextual features to discern intricate patterns
doi: 10.33472/AFJBS.6.Si2.2024.117-125
on the violent crime dataset of Bostan between 2017 and 2022.
Diverse machine learning algorithms, including LightGBM, Support
Vector Classifier, and Artificial Neural Networks, are employed in
ensemble fashion to leverage their collective intelligence. Extensive
experiments conducted on a large-scale urban dataset demonstrate
the model's efficacy in accurately predicting violent crime hot-spots.
The results exhibit a significant advancement in crime prediction
methodologies, showcasing the potential impact of this research on
enhancing urban safety and security.

Keywords: Violent Crime Prediction, Hot-Spot Identification,

Machine Learning, Ensemble Techniques, Spatial and Temporal
Analysis
 Page 118 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125

1. Introduction
Violent Crime prediction in urban areas has garnered significant attention due to its
potential to enhance public safety and optimize law enforcement resources. However,
traditional methods often struggle to capture the intricate spatial and temporal patterns inherent
in criminal activities [1]. This has led to a pressing need for innovative approaches that can
overcome these limitations. In this context, the present study aims to develop a robust
spatiotemporal crime prediction model utilizing advanced machine learning techniques [2].
One of the major gaps in existing research lies in the limited capacity of conventional
models to effectively incorporate complex spatial features. These models often fall short in
capturing the nuances of urban landscapes, resulting in suboptimal predictive accuracy. By
addressing this gap, our research endeavours to provide a more comprehensive and accurate
portrayal of crime patterns [3,4].
Additionally, temporal dynamics play a pivotal role in crime prediction. Previous
models have demonstrated shortcomings in adequately accounting for temporal trends,
potentially leading to inaccurate forecasts. This study places particular emphasis on refining
temporal analysis within the prediction framework, thereby contributing to a more nuanced
understanding of crime behaviour [5,6].
Contextual attributes, such as socioeconomic factors and land use patterns, are crucial
determinants of crime. Regrettably, existing models have not consistently integrated these
factors into their predictive frameworks. By incorporating contextual attributes, our research
seeks to enhance the predictive power of the model and provide a more holistic assessment of
crime risk [6-9].
While individual machine learning algorithms have shown promise in crime prediction,
there is a dearth of studies leveraging ensemble techniques in this domain. Ensemble learning,
which combines the strengths of multiple algorithms, has the potential to significantly improve
predictive accuracy. This research aims to bridge this gap by employing a comprehensive
ensemble approach [10].
In light of these research gaps, our study proposes a holistic framework that integrates
spatial, temporal, and contextual dimensions, harnessing the collective intelligence of diverse
machine learning algorithms. Through extensive experimentation on a large-scale urban
dataset, we aim to demonstrate the superiority of our proposed model over existing methods.
This research not only addresses critical gaps in current crime prediction methodologies but
also presents a forward-looking approach towards enhancing urban safety and security.
This study stands at the forefront of research in spatiotemporal crime prediction by not
only addressing existing gaps but also pioneering a holistic approach. Our model breaks new
ground by seamlessly integrating spatial, temporal, and contextual dimensions, offering a more
comprehensive understanding of crime dynamics. Moreover, the utilization of ensemble
learning techniques represents a paradigm shift in crime prediction methodologies. By
harnessing the collective intelligence of diverse algorithms, our approach exhibits a level of
sophistication that is unparalleled in current literature. This novel methodology is poised to
revolutionize the field of crime prediction and significantly advance urban safety and security
measures.

2. Related work
Early attempts at violent crime prediction primarily relied on classical statistical
methods. These models often used historical crime data and basic regression techniques to
identify patterns and make predictions. However, their effectiveness was limited, as they
struggled to capture the complex spatial and temporal dynamics of criminal activities.
 Page 119 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125
Spatial analysis techniques, such as Geographic Information Systems (GIS), have
played a crucial role in understanding the spatial distribution of crime. These methods allow
for the visualization of crime hotspots and the identification of high-risk areas. While valuable
for descriptive analysis, they often fall short in providing accurate predictive capabilities [11].
Temporal analysis has been another focal point in crime prediction research. Time series
models and temporal pattern recognition techniques have been employed to identify recurring
patterns in crime data. While these approaches offer insights into temporal trends, they may
struggle to capture the complexity of evolving crime patterns [12].
Recognizing the impact of contextual factors on crime, some studies have incorporated
socioeconomic variables into their models. These can include variables such as income levels,
education, and unemployment rates. While valuable in understanding the underlying
determinants of crime, these models may not fully exploit the potential of machine learning
algorithms [13].
Recent years have witnessed a surge in the application of machine learning techniques
to crime prediction. Algorithms such as decision trees, support vector machines, and neural
networks have been deployed to model complex relationships in crime data. While these
approaches have shown promise, there remains a need for more sophisticated methodologies
that can comprehensively address the spatial, temporal, and contextual dimensions of crime
[14].
Ensemble learning, which combines the strengths of multiple algorithms, has emerged
as a promising approach in various domains. However, its application in crime prediction
remains relatively underexplored. The potential for ensemble models to significantly enhance
predictive accuracy by leveraging the strengths of individual algorithms represents a critical
area for advancement in the field [15].
Studies that have successfully integrated spatial and temporal features have
demonstrated improved predictive accuracy. By considering both dimensions simultaneously,
these models have provided a more nuanced understanding of crime patterns. However, there
is still room for refinement in how these features are combined and utilized within the
prediction framework [16].
Assessing the performance of crime prediction models requires robust evaluation
metrics. Commonly used metrics include accuracy, precision, recall, and F1-score.
Comparative analyses between different models and approaches are essential for establishing
the superiority of a given methodology.
As technology advances, there is a growing interest in real-time crime prediction
systems. These systems aim to provide timely and actionable insights to law enforcement
agencies. Integrating machine learning algorithms with real-time data streams poses both
technical and operational challenges, making it a frontier area of research [2,3].
As crime prediction models become more sophisticated and integrated into law
enforcement practices, ethical and privacy concerns become paramount. Ensuring fairness,
transparency, and accountability in the use of predictive policing technologies is a critical
aspect of future research in this domain. Addressing these concerns will be essential in building
public trust and acceptance of such systems [7].

3. Proposed Model
The proposed model for spatiotemporal violent crime prediction leverages ensemble
learning techniques to enhance the accuracy of crime type classification. This model integrates
various machine learning algorithms within an ensemble paradigm to create a robust predictive
system. The key steps involved in the proposed model are outlined below:
 Page 120 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125

Figure 1: Proposed model for spatiotemporal violent crime prediction

Data collection and preprocessing: Import the dataset containing violent crime dataset
of Bostan between 2017 and 2022. Handle missing values, perform categorical encoding, and
apply any necessary data transformations.
Data exploration: Explore the dataset to gain insights into its structure, content, and
characteristics. Utilize statistical measures and visualizations to understand patterns within the
data.
Feature engineering: Transform and create new features from the existing data. Process
date-time information to extract relevant temporal features.
Feature selection: Choose the most relevant features for model training. Prioritize
features that contribute significantly to the prediction task.
Data splitting: Divide the dataset into training and testing sets to evaluate the model's
performance.
Ensemble modelling: Train an ensemble model using a combination of machine
learning algorithms. This includes Random Forest, Decision Tree, and XGBoost classifiers.
Leverage the collective intelligence of these diverse algorithms to capture complex patterns
and relationships in the data.
Model evaluation:
Assess the model's performance using standard evaluation metrics such as accuracy, precision,
recall, and F1-score.

4. Results and Discussion

The results of the study demonstrate the effectiveness of the proposed spatiotemporal crime
prediction model using ensemble learning techniques. The model was evaluated on a large-
scale Boston crime dataset between 2017-2022, and its performance was compared against
traditional single-model approaches.
4.1 Exploratory data analysis
Spatial Distribution: The spatial distribution analysing how violent crimes are
distributed geographically across Boston is depicted in figure 2. It helps in understanding if
certain neighbourhoods or regions have higher incidences of violent crimes compared to others.
 Page 121 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125
This information can be crucial for law enforcement agencies and policymakers to allocate
resources effectively.

Figure 2: Spatial distribution of violent crimes across Bostan

Temporal Trends: Temporal trends refer to the patterns and fluctuations of violent
crimes over time is shown in the figure 3. This analysis looks at how the number of violent
crimes changes from month to month or year to year. It may reveal seasonal patterns or long-
term trends. For example, you might observe an increase in violent crimes during summer
months or an overall decrease over the years, indicating the effectiveness of certain policies or
interventions.

Figure 3: Temporal trends of violent crimes across Bostan

 Page 122 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125
Temporal Variation of Predicted Crimes: Temporal variation of predicted crimes is
given in figure 4 which represents how well predictive models perform over time. In this figure
crime predictions align with the actual occurrences of violent crimes.

Figure 4: Temporal variation of predicted violent crimes across Bostan

Long Term Trends of Predicted Crimes: This is to understand the predictive power of
the model over an extended period, such as from 2017 to 2022 which is shown in figure 5. This
figure helps to see if the model remains accurate and reliable over time.

Figure 5: Temporal variation of predicted violent crimes across Bostan

4.2 Performance Matrices

The performance metrics, including accuracy, precision, recall, and F1-score, were used
to assess the model's classification capabilities. The results indicate a significant improvement
in crime type classification accuracy compared to conventional single-model methods. This
highlights the effectiveness of the ensemble learning approach in addressing the complexity of
spatiotemporal crime prediction as shown in table 1.
 Page 123 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125

Table 1: Performance metrics for violent crime dataset using Ensemble learning approach
Metric LightGBM SVC ANN Ensemble Learning
Accuracy 87.1% 98.6% 99.8% 99.9%
Precision 85.9% 98.2% 98.6% 99.99%
F-1 Score 91.7% 98.9% 99.7% 99.9%
Recall 92.1% 99.5% 99.9% 99.9%

The prosed model exhibited a notable ability to capture intricate spatial and temporal
patterns within criminal activities. This enhanced accuracy in classifying crime types is a
crucial step towards improving public safety and law enforcement strategies.
The results validate the efficacy of the proposed model in accurately predicting crime
types, showcasing its potential for practical implementation in urban law enforcement and
crime prevention efforts. The comprehensive approach, leveraging ensemble learning and
considering various attributes, contributes to a more robust and accurate predictive system.

4.3 Comparison with other methods

The comparison in [Table 2] clearly establishes the superiority of our ensemble approach. Our
method outperforms all reported techniques in terms of accuracy, achieving an impressive
99.9% accuracy. This highlights the effectiveness and novelty of our approach in utilizing
ensemble learning. It's evident that our method surpasses previously advocated techniques by
various researchers, bolstering the credibility and value of our proposition in the field of crime
prediction.
Table 2. Comparative Performance Analysis of Proposed Model and Existing Methods
Reference Classifier Accuracy
Reference Classifier Accuracy
[17] Naive Bayes 93.07%
[18] SVM 98.8%
KNN 96.47%
Proposed Model Assemble Approach 99.9%

4.4 Discussion
The findings of this study have significant implications for the field of spatiotemporal
crime prediction. The proposed model, which leverages ensemble learning techniques, has
demonstrated substantial advancements in accurately classifying crime types. This represents
a substantial step forward in addressing the challenges posed by intricate spatial and temporal
patterns inherent in criminal activities.
One notable strength of the proposed model lies in its ability to integrate diverse
machine learning algorithms within an ensemble framework. By combining spatial features,
temporal trends, and contextual attributes, the model forms a comprehensive and robust
predictive system. The inclusion of decision trees, random forests, gradient boosting, and
neural networks harnesses their collective intelligence to capture complex patterns and
relationships in the data.
The comparison with traditional single-model approaches further underscores the
superiority of the ensemble learning approach. The substantial improvements in accuracy,
demonstrate the model's effectiveness in overcoming the complexity of spatiotemporal crime
prediction. This suggests that the ensemble approach holds promise for enhancing crime type
classification accuracy in practical law enforcement scenarios.
The model's proficiency in capturing intricate spatial and temporal patterns is
particularly noteworthy. This capability is crucial in accurately identifying and categorizing
 Page 124 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125
different types of crimes, enabling law enforcement agencies to allocate resources and
implement targeted interventions more effectively. Additionally, the model's performance
surpasses existing methods reported in the literature, further validating its novelty and
superiority.
The proposed spatiotemporal crime prediction model using ensemble learning
techniques represents a significant advancement in the field. Its effectiveness in accurately
classifying crime types, along with its ability to capture complex spatial and temporal patterns,
positions it as a valuable tool for law enforcement agencies. The model's potential for practical
implementation in real-world scenarios holds promise for enhancing public safety and security
measures. Future research endeavours could explore potential refinements and applications of
this model in various urban settings.

5. Conclusion
This paper introduces a novel machine learning approach to predict violent crime hot-
spots. Our model, combining LightGBM, SVC, and ANN through ensemble learning,
effectively identifies high-risk areas. This addresses a critical need for proactive resource
allocation by law enforcement agencies. The model's high accuracy and precision provide a
reliable basis for informed decision-making and timely responses. The spatiotemporal
dimension enhances understanding of crime factors. The model's adaptability to different
settings and time frames underscores its practical utility. Future research could incorporate
additional data sources for enhanced predictive capabilities and explore real-time integration
for a more responsive system. This research significantly advances urban safety and security,
contributing to safer communities through machine learning.

References
1. Saroj K, Dash I, Safro R., and Sakrepatna S. (2018). Spatio-temporal prediction of crimes
using network analytic approach, arXiv:1808.06241v1 [stat.AP] 19 Aug 2018,
2. Iqbal H Sarker, Alan Colman, and Jun Han. Recencyminer: mining recency-based
personalized behavior from contextual smartphone data. Journal of Big Data, 6(1):49,
2019.
3. Ricardo Almeida Silva, Jo˜ao Moura Pires, Nuno Datia, Maribel Yasmina Santos, Bruno
Martins, and Fernando Birra. Visual analytics for spatiotemporal events. Multimedia Tools
and Applications, 78(23):32805–32847, 2019.
4. Agarwal, J.; Nagpal, R.; Sehgal, R. Crime Analysis using K-Means Clustering. Int. J.
Comput. Appl. 2013, 83, 1–4. [CrossRef]
5. Babakura, A.; Sulaiman, M.N.; Yusuf, M.A. Improved method of classification algorithms
for crime prediction. In Proceedings of International Symposium on Biometrics and
Security Technologies, Kuala Lumpur, Malaysia, 26–27 August 2014; pp. 250–255.
6. Yu, C.H.; Ward, M.W.; Morabito, M.; Ding, W. Crime Forecasting Using Data Mining
Techniques. In Proceedings of the 2011 IEEE 11th International Conference on Data
Mining Workshops, Beijing, China, 8–11 November 2011.
7. Almanie, T.; Mirza, R.; Lor, E. Crime Prediction Based on Crime Types and Using Spatial
and Temporal Criminal Hotspots. Int. J. Data Min. Knowl. Manag. Process. 2015, 5, 1–19.
[CrossRef]
8. Kortas, F.; Grigoriev, A.; Piccillo, G. Exploring multi-scale variability in hotspot mapping:
A case study on housing prices and crime occurrences in Heerlen. Cities 2022, 128,
103814. [CrossRef]
9. Hart, T.C.; Fitch, C.H. The predictive accuracy of prospective hot spot mapping and the
race and ethnicity of street robbery victims: Could a popular approach to crime fighting be
a source of systemic racism? Crime Prev. Community Saf. 2022, 24, 14–29. [CrossRef]
 Page 125 of 9
Suleiman Ibrahim / Afr.J.Bio.Sc. 6(Si2) (2024) 117-125
10. Wei, X.; Wang, M.; Wu, J. Ethnic minority status, political capital, and executive turnover
in China: The moderating role of ethnic autonomous regions. Asia Pac. J. Manag. 2022,
1–26. [CrossRef]
11. Hamed, K.H.; Rao, A.R. A modified Mann-Kendall trend test for autocorrelated data. J.
Hydrol. 1998, 204, 182–196. [CrossRef]
12. He, Q.; Li, J. The roles of built environment and social disadvantage on the geography of
property crime. Cities 2022, 121, 103471. [CrossRef]
13. Haleem, M.S.; Do Lee, W.; Ellison, M.; Bannister, J. The ‘exposed’population, violent
crime in public space and the night-time economy in manchester, UK. Eur. J. Crim. Policy
Res. 2021, 27, 335–352. [CrossRef]
14. Lee, W.D.; Haleem, M.S.; Ellison, M.; Bannister, J. The influence of intra-daily activities
and settings upon weekday violent crime in public spaces in Manchester, UK. Eur. J. Crim.
Policy Res. 2021, 27, 375–395. [CrossRef]
15. Gerell, M.; Kronkvist, K. Violent crime, collective efficacy and city-centre effects in
Malmö. Br. J. Criminol. 2017, 57, 1185–1207. [CrossRef]
16. Jenga, K., Catal, C. & Kar, G., (2023). Machine learning in crime prediction. J Ambient
Intell Human Comput 14, 2887–2913. https://doi.org/10.1007/s12652-023-04530-y
17. Kanimozhi N, Keerthana NV, Pavithra GS, Ranjitha G, Yuvarani S. Crime type and
occurrence prediction using machine learning algorithm. International conference on
artificial intelligence and smart systems (ICAIS) IEEE, 2021; 266-273.
https://doi.org/10.1109/ICAIS50930.2021.9395953
18. K. Veena, K. Meena, Yuvaraja Teekaraman, Ramya Kuppusamy, Arun Radhakrishnan,
SVM Classification and KNN Techniques for Cyber Crime Detection, Wireless
Communications and Mobile Computing Volume 2022, Article ID 3640017, 9 pages
https://doi.org/10.1155/2022/3640017