Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

Forecasting Criminal Activity Using Machine

Learning Approaches
1st M. Vasuki ; 2nd Dr.T.Amalraj Victoire ; 3rd S.Seventhi
1
Associate professor, Department of Master Computer Applications, Sri Manakula Vinayagar Engineering Collage,
Puducherry – 605107, India.
2
Associate professor, Department of Master Computer Applications, Sri Manakula Vinayagar Engineering Collage,
Puducherry – 605107, India.
3
Student, Department of Master Computer Applications, Sri Manakula Vinayagar Engineering Collage, Puducherry – 605107, India.

Abstract:- Predicting criminal activity has long been a recent times has presented encouraging prospects for
challenge for law enforcement agencies worldwide. enhancing the capacity to forecast crimes. Machine learning
Traditional methods often rely on historical data and models have the capacity to reveal intricate patterns and
human intuition, which may be limited in their accuracy linkages in crime data that may not be readily discernible to
and scope. In recent years, machine learning techniques human analysts, thanks to their ability to utilize extensive
have emerged as promising tools for forecasting criminal datasets and advanced algorithms.
activity by leveraging large-scale datasets and advanced
algorithms. This paper presents a novel machine learning These models have the potential to enhance law
approach to forecasting criminal activity, focusing on the enforcement efforts by providing timely and actionable
development and evaluation of predictive models using insights into emerging criminal trends, enabling proactive
various data sources, including crime reports, interventions and resource allocation. This paper aims to
demographic information, and environmental factors. We explore the application of machine learning approaches to
explore the application of supervised and unsupervised forecasting criminal activity, with a focus on developing
learning algorithms, such as decision trees, random forests, accurate and interpretable predictive models. We will examine
support vector machines, and neural networks, to identify various machine learning algorithms, including supervised and
patterns and trends in crime data. Furthermore, we unsupervised learning techniques, and assess their
discuss the challenges and ethical considerations associated performance in predicting different types of criminal offenses
with deploying predictive models in real-world law across diverse geographical and temporal contexts.
enforcement settings, emphasizing the importance of
transparency, fairness, and accountability. Through We will also go over the difficulties and moral questions
empirical analysis and case studies, we demonstrate the pertaining to the use of predictive policing technologies, such
potential of machine learning techniques to enhance crime as those involving data privacy, bias, and accountability. Even
prediction and prevention efforts, providing valuable while machine learning has a lot of potential to improve
insights for policymakers, law enforcement agencies, and efforts at crime prediction and prevention, it is crucial to make
researchers in the field of criminal justice. sure that these technologies are applied in a way that respects
individual rights, justice, and transparency. In order to further
Keywords:- Predicting Criminal, Machine Learning the ongoing conversation on utilizing data-driven approaches
Techniques, Crime Reports, Demographic Information, to enhance public safety and security, this study aims to
Decision Trees, Random Forests, Enforcement Agencies. deepen our understanding of how machine learning might be
applied to crime forecasting.
I. INTRODUCTION
II. LITERATURE REVIEW
Predicting and preventing crimes have always been
difficult tasks for law enforcement organizations around the A thorough analysis of the body of research offers
globe. It is still difficult to anticipate and mitigate criminal insightful information about the state-of-the-art approaches,
activities successfully, despite advancements in technology difficulties, and methods for predicting criminal activity using
and policing techniques. Conventional approaches to crime machine learning techniques. A vast array of research projects,
forecasting frequently depend on human judgment and the scholarly articles, and real-world applications in the fields of
study of past data, which may have limitations in terms of crime forecasting and predictive policing are included in the
their prediction precision and capacity to adjust to changing literature study. The main ideas and conclusions drawn from
crime trends. The advent of machine learning techniques in the literature review are outlined below:

IJISRT24MAY645 www.ijisrt.com 732

Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

 Mohler et al. (2015), "Predictive Policing: The Role of evaluation methodologies to enhance the effectiveness of
Crime Forecasting in Law Enforcement Operations": The predictive policing efforts.
use of machine learning techniques to forecast crime
hotspots and patterns is covered in this research. It includes B. Features and Modules
a variety of methods and assesses how effective they are at
predicting crimes, including random forests, support vector  Data Collection and Preprocessing: The first step in our
machines, and neural networks. system involves collecting diverse datasets relevant to
 "Crime Prediction using Machine Learning Techniques" by crime prediction, including historical crime records,
Malathi and Geetha (2018): This study explores the use of demographic information, socio-economic indicators,
machine learning algorithms like decision trees, k-nearest environmental factors, and geographical features. These
neighbors, and logistic regression for crime prediction. It datasets are preprocessed to handle missing values,
discusses the challenges and limitations of existing outliers, and inconsistencies, ensuring the quality and
approaches and proposes novel techniques to improve integrity of the data for model training.
prediction accuracy.  Feature Engineering and Selection: Next, we perform
 "A Review of Crime Prediction Methods" by Amorim et al. feature engineering to extract meaningful patterns and
(2020): This review paper provides an overview of relationships from the raw data. This includes transforming
different machine learning methods used in crime categorical variables, creating new features, and selecting
prediction, including traditional statistical models and relevant attributes for model training. Feature selection
more advanced techniques like deep learning. It discusses techniques such as correlation analysis, principal
the strengths and weaknesses of each approach and component analysis (PCA), and recursive feature
identifies areas for future research. elimination (RFE) are employed to identify the most
 "Crime Prediction Using Machine Learning Algorithms: A informative predictors for crime forecasting.
Survey" by Ahmed et al. (2019): This survey paper  Model Development: Our system utilizes a variety of
summarizes recent advancements in crime prediction using machine learning algorithms, including supervised and
machine learning algorithms. It categorizes existing unsupervised learning techniques, to build predictive
approaches based on the types of crime prediction tasks models. These algorithms include decision trees, random
(e.g., hotspot prediction, crime type classification) and forests, support vector machines (SVM), neural networks,
provides insights into the key factors influencing and ensemble methods. We explore the strengths and
prediction accuracy. weaknesses of each algorithm and assess their performance
 "Predictive Policing: Review and Evaluation of Machine in terms of predictive accuracy, scalability, and
Learning Models for Crime Prediction" by Borrion and interpretability.
Cherney (2018): This paper reviews the application of  Evaluation and Validation: The performance of the
machine learning models in predictive policing and predictive models is evaluated using rigorous validation
evaluates their performance based on various metrics such methodologies, such as cross-validation, holdout
as accuracy, precision, and recall. It also discusses ethical validation, and temporal validation. We assess the models'
and privacy concerns associated with the use of predictive predictive accuracy, precision, recall, F1-score, and area
policing technologies. under the receiver operating characteristic (ROC) curve to
determine their effectiveness in forecasting different types
III. EXISTING SYSTEM of criminal activity across various spatio-temporal
contexts.
In researchers and law enforcement agencies have  Deployment and Integration: Once validated, the
increasingly turned to machine learning techniques to enhance predictive models are deployed within a user-friendly
the accuracy and efficiency of crime forecasting. Several interface accessible to law enforcement agencies and
existing systems and methodologies have been developed to policymakers. The system provides real-time crime
address the complex challenges associated with predicting forecasts, interactive visualizations, and actionable insights
criminal activity. These systems typically leverage historical to support decision-making and resource allocation.
crime data, demographic information, geographic features, and Integration with existing law enforcement systems and
environmental factors to train predictive models. workflows ensures seamless adoption and utilization of the
predictive policing technology.
A. Proposed System  Ethical Considerations and Transparency: Throughout the
In response to the growing demand for more accurate development and deployment process, we prioritize ethical
and efficient crime prediction tools, we propose a considerations, including fairness, transparency, and
comprehensive system leveraging machine learning accountability. Measures are implemented to mitigate
techniques to forecast criminal activity. Our proposed system biases, ensure data privacy and security, and foster
integrates state-of-the-art algorithms, data sources, and community engagement and trust. Model interpretability
techniques, such as feature importance analysis and model

IJISRT24MAY645 www.ijisrt.com 733

Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

explanations, are incorporated to enhance transparency and C. Architecture Design

facilitate human oversight of the forecasting process.

Fig.1 Architecture of Data Processing

D. Crime Prediction Techniques  Predictive Modeling: Predictive modeling involves

building statistical or machine learning models to predict
Crime prediction techniques encompass a variety of future crime occurrences based on various predictors or
methodologies used to forecast criminal activity, ranging from features. Supervised learning techniques such as decision
traditional statistical approaches to advanced machine learning trees, random forests, support vector machines (SVM),
algorithms. Here are some commonly employed crime logistic regression, and neural networks are commonly
prediction techniques: used for crime prediction. Unsupervised learning
 Time Series Analysis: Time series analysis involves techniques such as clustering algorithms (e.g., k-means,
analyzing historical crime data to identify patterns and DBSCAN) are also employed for anomaly detection and
trends over time. Techniques such as autoregressive pattern recognition.
integrated moving average (ARIMA) models, seasonal  Risk Assessment Tools: Risk assessment tools are used to
decomposition, and exponential smoothing are used to evaluate the likelihood of individuals reoffending or
forecast future crime occurrences based on past data. committing future crimes based on their demographic
 Spatial Analysis: Spatial analysis focuses on the characteristics, criminal history, and other risk factors.
geographical distribution of crime incidents and the These tools, such as the Correctional Offender
identification of crime hotspots. Spatial techniques such as Management Profiling for Alternative Sanctions
kernel density estimation (KDE), spatial autocorrelation (COMPAS), employ machine learning algorithms to assess
analysis, and point pattern analysis are utilized to identify recidivism risk and inform sentencing decisions in the
high-risk areas and allocate resources effectively. criminal.

IJISRT24MAY645 www.ijisrt.com 734

Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

IV. EXPERIMENTAL RESULTS

 Data Set

Fig. 2 Crime Analysis Performance

Fig. 3 Victim Report of Varience

IJISRT24MAY645 www.ijisrt.com 735

Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

 Random Forest crime prediction jobs because of their ease of use,

In crime prediction using machine learning, Random interpretability, and capacity to produce useful information for
Forest stands out as a formidable tool for its ability to handle resource allocation and crime prevention tactics.
complex datasets and provide accurate predictions. By
leveraging historical crime data along with various socio-  Support Vector Machine
economic, demographic, and geographic factors, Random Support Vector Machines (SVMs) efficiently distinguish
Forest models can effectively identify patterns and between various crime classes in large, complicated datasets,
relationships that contribute to criminal activity in specific providing a reliable method for machine learning crime
areas. In this context, Random Forest algorithms can analyze prediction. SVMs function by maximizing the margin between
features such as time of day, day of the week, location, two classes by identifying the hyperplane that optimally splits
weather conditions, population density, unemployment rates, the dataset into those classes. To classify places or populations
and proximity to social amenities or crime hotspots. By into distinct risk categories in the context of crime prediction,
training on labeled data where crimes are categorized by type support vector machines (SVMs) examine a variety of
and location, Random Forest models can learn to predict the parameters, including time, location, demographics, socio-
likelihood of different types of crimes occurring in various economic indicators, and past crime data. SVMs can reliably
areas. Moreover, the algorithm's ability to assess feature classify new occurrences by locating the ideal hyperplane,
importance aids law enforcement agencies and policymakers indicating whether a specific location is more likely to see a
in understanding the key factors driving crime rates, thereby particular sort of crime. For instance, an SVM may find
enabling targeted interventions and resource allocation. temporal and spatial patterns suggesting a higher risk of
However, it's crucial to ensure the ethical and responsible use violent crimes in crowded cities at specific times of the day.
of such predictive models, considering potential biases in data
collection and algorithmic predictions that could Furthermore, by utilizing kernel functions, SVMs can
disproportionately impact certain communities. Additionally, handle high dimensionality datasets and non-linear feature
continuous evaluation and refinement of the model based on associations, facilitating the investigation of intricate variable
new data and feedback are essential to maintain its interactions. Nevertheless, selecting the right kernel and
effectiveness and relevance in the dynamic landscape of crime hyperparameters can have a significant impact on SVM
prevention and public safety. performance, necessitating careful tuning. In spite of this,
SVMs have a number of benefits when it comes to crime
 Decision Tree prediction tasks, including as their resilience to overfitting and
Decision trees are a fundamental tool in machine their capacity to handle imbalanced datasets. Through precise
learning-based crime prediction because of their forecasts and analyses of crime hotspots and trends, SVMs
interpretability and capacity to reveal intricate patterns in enable law enforcement organizations and decision-makers to
crime data. With each decision node representing a feature and efficiently allocate resources, carry out focused interventions,
each leaf node corresponding to a predicted outcome, decision and improve community safety.
trees divide the feature space based on a sequence of if-else
criteria. Decision trees examine a number of variables, V. FUTURE WORK
including time, place, demographics, socioeconomic
indicators, and past crime data, in the context of crime Building on the findings and insights gained from the
prediction. Decision trees are able to determine important current study, future research directions can be identified to
predictors of criminal activity in particular regions or further advance the field of forecasting criminal activity using
communities by recursively dividing the dataset into subsets a machine learning approach. Future research can explore the
based on these characteristics. Decision trees, for example, can integration of additional data sources, such as social media
show that communities with high unemployment rates and data, surveillance footage, and sensor data, to enhance the
poor lighting are more likely to experience property crimes at predictive capabilities of the models. By incorporating diverse
night. data modalities, researchers can capture a more
comprehensive understanding of the underlying factors
Decision trees also provide transparency by providing a influencing criminal behavior. There is a need for continued
visual representation of the decision-making process, which research into developing techniques for enhancing the
helps politicians and law enforcement organizations interpretability and transparency of predictive models.
comprehend the variables influencing crime trends. Decision
trees, however, are prone to overfitting, a phenomenon in Future work can focus on refining existing methods for
which noise in the training data gets captured by the model, feature importance analysis, model explanations, and
impairing its ability to generalize to new data. This problem visualizations to facilitate human understanding and oversight
can be lessened by employing ensemble techniques like of the forecasting process. Addressing algorithmic bias and
Random Forest, pruning, and restricting the depth of trees. fairness concerns remains a critical area for future research.
Decision trees have limits, but they are nonetheless useful in Researchers can investigate fairness-aware learning techniques

IJISRT24MAY645 www.ijisrt.com 736

Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

and algorithms that explicitly account for biases in the data  Disadvantages
and mitigate discriminatory outcomes. This includes exploring While utilizing machine learning techniques to foresee
methods for detecting and correcting biases in training data criminal activity has several benefits, there are a number of
and designing algorithms that prioritize fairness and equity in drawbacks and issues that must be resolved.
predictions. Future research should prioritize community  Discrimination and Bias: Machine learning models trained
engagement and collaboration in the development and on past crime data may reinforce discriminatory and biased
deployment of predictive policing technologies. tendencies seen in the data, producing unfair or
discriminating results. Predictive biases can be produced
Researchers can work closely with stakeholders, by training data biases, such as the overrepresentation of
including law enforcement agencies, policymakers, civil rights particular neighborhoods or demographics, which might
organizations, and community members, to co-design and worsen already-existing inequalities in law enforcement
evaluate predictive policing systems that reflect community procedures.
values, preferences, and concerns. Longitudinal studies and  Availability and Quality of Data: There can be substantial
impact evaluations are needed to assess the long-term differences in the availability and quality of crime statistics
effectiveness and societal impact of predictive policing between different jurisdictions and historical periods.
technologies. Future research can focus on conducting Machine learning models can be rendered unreliable and
rigorous evaluations of predictive policing systems in real- their efficacy compromised by incomplete, inaccurate, or
world settings, measuring their impact on crime rates, police- biased data. Furthermore, utilizing and incorporating
community relations, and social outcomes over time. information from a variety of sources, including social
media
 Advantages
Forecasting criminal activity using a machine learning VI. CONCLUSION
approach offers several advantages that can significantly
enhance law enforcement efforts and public safety. In conclusion, forecasting criminal activity using
 Early Detection of Crime Trends: Machine learning machine learning approaches represents a promising avenue
models can analyze large volumes of historical crime data for enhancing public safety and improving law enforcement
and identify emerging patterns and trends that may not be strategies. Crime prediction is a multifaceted endeavor that
immediately apparent to human analysts. By detecting hinges on the interplay between advanced algorithms,
crime trends early, law enforcement agencies can meticulous data analysis, and domain expertise. This study has
proactively allocate resources and implement targeted explored the application of machine learning algorithms, such
interventions to prevent future criminal activity. as random deep learning model, graph-based model, Ensemble
 Prediction Accuracy: Machine learning algorithms, such as learning to predict future crime occurrences based on
random forests, support vector machines, and neural historical data and various contextual factors. While the use of
networks, are capable of capturing complex relationships machine learning in crime forecasting offers several
and patterns in crime data, leading to more accurate advantages, including early detection of crime trends,
predictions of future criminal activity. These models can prediction accuracy, and resource optimization, it also presents
consider multiple factors, including geographical, significant challenges and limitations. One of the key
temporal, and socio-economic variables, to generate challenges is the potential for bias and discrimination in
precise forecasts of crime occurrences. predictive models, as they may perpetuate existing biases
 Resource Optimization: By accurately forecasting criminal present in the training data. Addressing bias requires careful
activity, law enforcement agencies can optimize the attention to data quality, algorithmic fairness, and ethical
allocation of resources, such as patrol officers, considerations to ensure that predictive policing technologies
investigative units, and crime prevention initiatives. are used responsibly and equitably. Additionally, concerns
Predictive models can identify high-risk areas and times, related to data privacy, interpretability, and unintended
allowing agencies to deploy personnel and resources consequences must be addressed to build trust and
strategically to deter criminal behavior and respond accountability in predictive policing systems.
promptly to incidents.
 Proactive Crime Prevention: Machine learning-based REFERENCES
crime prediction enables law enforcement agencies to
adopt a proactive approach to crime prevention. By [1]. Mohler, G. O., Short, M. B., Malinowski, S., Johnson,
anticipating where and when crimes are likely to occur, M., Tita, G. E., Bertozzi, A. L., & Brantingham, P. J.
agencies can implement preventive measures, such as (2015). Randomized controlled field trials of predictive
increased patrols, community outreach programs, and policing. Journal of the American Statistical Association,
targeted interventions, to disrupt criminal activity and 110(512), 1399-1411.
improve public safety.

IJISRT24MAY645 www.ijisrt.com 737

Volume 9, Issue 5, May – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAY645

[2]. Santos, R. L., Oliveira, T. A., Santos, R. M., & Prates, M.

O. (2018). Crime prediction through urban metrics and
statistical learning techniques. Expert Systems with
Applications, 113, 83-92.
[3]. Mohler, G. O., Short, M. B., Malinowski, S., Johnson,
M., Tita, G. E., Bertozzi, A. L., & Brantingham, P. J.
(2011). A model for forecasting the locations of crimes.
Journal of the American Statistical Association,
106(496), 147-157.
[4]. Ribeiro, F. N., Santos, J. C., & Oliveira, D. N. (2016).
Crime prediction through urban metrics and statistical
learning techniques. Expert Systems with Applications,
46, 254-261.
[5]. Caplan, J. M., Kennedy, L. W., & Piza, E. L. (2011). Join
the utility of event-dependent of the crime analysis
techniques for violent crime forecasting. Crime &
Delinquency, 57(2), 192-214.
[6]. Piza, E. L., Feng, S., & Kennedy, L. W. (2017). Short-
term forecasting of violent crime: Application to
Indianapolis. Journal of Quantitative Criminology, 33(3),
467-488.
[7]. Gerber, M. S., & Johnson, B. D. (2010). Using spatial
metrics to predict part I offenses in a medium-sized city.
Criminal Justice Review, 35(1), 89-111.
[8]. Ashby, M. P. J., & Bowers, K. J. (2013). The (in)
effectiveness of predictive policing: A review and
critique. Crime Science, 2(1), 1-12.
[9]. Borrion, H., & Muttaqien, A. (2020). The prospects of
predictive policing: A systematic review. Crime Science,
9(1), 1-18.
[10]. Drawve, G., & Taylor, R. B. (2019). Does predictive
policing lead to biased arrests? A comparative analysis of
Chicago and San Francisco. Crime & Delinquency,
65(5), 569-601.
[11]. Kadar, C. S., Shahid, S., Shafiq, Z., & Malik, M. S.
(2019). Crime prediction and analysis using machine
learning. International Journal of Scientific &
Technology Research, 8(11), 4402-4406.
[12]. Wang, H., & Li, S. (2020). Crime prediction using
machine learning algorithms. In International Conference
on Artificial Intelligence in Information and
Communication (pp. 131-138). Springer, Singapore.
[13]. Tavana, M., & Ha, T. (2017). Crime prediction using data
mining techniques: A review. Expert Systems with
Applications, 75, 83-100.
[14]. Groff, E. R., & La Vigne, N. G. (2002). Forecasting the
future of predictive crime mapping. Crime Mapping, 4,
6-10.
[15]. Wu, C. H., & Chen, Y. C. (2019). A review on crime
prediction using machine learning techniques.
Computers, Materials & Continua, 59(2), 681-694.

IJISRT24MAY645 www.ijisrt.com 738