EAST POINT COLLEGE OF

ENGINEERING AND TECHNOLOGY

Department of Information Science and Engineering
Jnana Prabha, Bidarahalli , Virgo Nagar Post, Bengaluru,
Karnataka-560049
Technical Seminar(18CSS84)
A SEMINAR ON
“CRIME PREDICTION USING MACHINE LEARNING
AND DEEP LEARNING”
UNDER THE GUIDANCE OF: PRESENTED BY:
PROF.REBECCA KRITHIKA M
ASSISTANT PROFESSOR 1EP20IS046
DEPT OF ISE, EPCET
AGENDA

■ INTRODUCTION
■ LITERATURE SURVEY
■ METHODOLOGY
■ ARCHITECTURE FLOW
■ PROPOSED SYSTEM/IMPLEMENTATION
■ ADVANTAGE
■ DISADVANTAGE
■ APPLICATION
■ CONCLUSION
INTRODUCTION
■ The article discusses the recent advancements in using machine learning and deep learning for crime
prediction. It highlights how algorithms like decision trees and support vector machines analyze crime data to
predict patterns and uncover trends. These technologies aid in developing tailored crime prevention strategies
based on demographic and environmental factors. Predictive policing, enabled by machine learning, helps
identify crime hotspots and allocate resources effectively for prevention efforts.
■ Deep learning algorithms, such as convolutional and recurrent neural networks, show promise in predicting
crime patterns and analyzing surveillance footage to detect criminal activities. However, challenges like data
quality, ethical concerns, and model interpretability remain.
■ The article concludes by emphasizing the importance of ongoing research in leveraging these technologies for
crime prediction and suggests future directions to enhance neighborhood crime detection, contributing to
advancements in the field.
LITERATURE SURVEY

1. Paper 1: "Predictive Policing: The Role of Machine Learning and Data Analytics
in Law Enforcement" by Smith et al. (2018)
■ Smith et al. provide an in-depth exploration of predictive policing methods, focusing on
the utilization of machine learning and data analytics in law enforcement. The paper
discusses various machine learning algorithms such as decision trees, support vector
machines, and random forests employed for crime prediction tasks. It presents case
studies highlighting successful implementations of predictive policing models in
different urban settings. Additionally, the authors discuss challenges associated with
data quality, model interpretability, and ethical considerations in predictive policing
initiatives. The paper emphasizes the importance of collaboration between law
enforcement agencies, data scientists, and policymakers to address these challenges and
maximize the effectiveness of predictive policing efforts.
2. Paper 2: "Crime Prediction Using Deep Learning Techniques" by Zhang
and Li (2020)
■ Zhang and Li focus specifically on the application of deep learning techniques
for crime prediction. The paper reviews recent advancements in deep learning,
including convolutional neural networks (CNNs) and recurrent neural
networks (RNNs), and their effectiveness in analyzing crime data attributes
such as type, time, and location. The authors present case studies
demonstrating the superior predictive performance of deep learning models
compared to traditional machine learning algorithms in crime prediction tasks.
They also discuss challenges related to data preprocessing, model training,
and interpretability of deep learning models. The paper concludes by
suggesting future research directions to address these challenges and enhance
the application of deep learning techniques in crime prediction.
METHODOLOGY
■ The researchers aimed to explore efficient algorithms for predicting neighborhood crimes, building upon
prior work that utilized statistical analysis for crime prediction in New York City. Employing a systematic
approach, they selected papers for review from multiple databases related to crime prediction, using both
automated and manual search methods.
■ Initially, they constructed queries focusing on terms related to crime prediction, detection, machine
learning, and deep learning techniques for databases such as IEEE, Science Direct, and ACM. After
filtering out irrelevant studies, they selected 30 papers for inclusion in the main text and an additional 20
papers for the appendix based on relevance and novelty.
■ To avoid potential duplicates, the researchers focused on individual research libraries rather than utilizing
Google Scholar. They extracted research works using specific keywords from databases such as Science-
Direct Elsevier, ACM Digital Library, and IEEE Xplore Access Digital Library. Filters were applied to
include only full-text papers, and the search was further refined based on index terms used in the articles.
■ After gathering over 450 papers, they applied a filter on document titles, resulting in a total of 157 papers.
The authors then manually reviewed the papers, focusing on elements such as the article's focus or
objective, datasets utilized, algorithms applied, and reported accuracy rates. The primary aim of the survey
was to provide insights into various algorithms applied in crime prediction and inform the research
community about available datasets for applying novel algorithms and obtaining research results.
SELECTION METHODOLOGY
ARCHITECTURE FLOW
■ Crime prediction using machine and deep learning involves several key steps outlined in a process
that begins with data collection, gathering relevant data such as crime statistics, demographics,
and weather patterns. Data preprocessing follows, where the collected data is cleaned and
transformed into a usable format. The data is then split into training and testing sets for model
development and evaluation. Feature engineering selects relevant features from the data to
effectively train the model. Various machine and deep learning algorithms are applied to the data
for training and prediction purposes. Finally, trained models are evaluated using performance
metrics to assess accuracy and effectiveness in predicting crime and supporting decision-making
in law enforcement and crime prevention efforts.
■ Several datasets have been utilized in crime detection and prediction research, including the
Chicago Crime Dataset, London Crime Dataset, Los Angeles Crime Dataset, New York City
(NYC) Crime Dataset, and Philadelphia Crime Dataset. These datasets contain information on
crimes reported in their respective cities and have been used to create models predicting the
likelihood of specific types of crimes occurring in different areas. Additionally, global datasets
focusing on CCTV video footage, types of aggression, and weapons for real-time crime
predictions are utilized. These datasets offer valuable information for researchers to build crime
prediction models, aiding law enforcement agencies in preventing and responding to criminal
activities more effectively. The location and access to these datasets used by research articles
surveyed in this paper are listed for reference.
ARCHITECTURE FLOW OF CRIME PREDICTION
PROPOSED SYSTEM
■ Crime prediction using traditional machine learning techniques has proven effective, offering
various methods such as decision trees, support vector machines, logistic regression, and random
forests. These models analyze crime data to identify patterns for predicting criminal activity,
requiring fewer data points and being easier to interpret compared to deep learning approaches.
1) Machine Learning Based Regression Methods for Crime Prediction:
■ - Regression techniques have been employed to predict various crime scenarios, focusing on
prevalent crimes like motorcycle robbery and urban crimes. Techniques such as ARIMAX-TFM,
spatial analysis, and auto-regressive models have been utilized, achieving significant accuracy scores
in predicting crime trends in specific regions.
■ - Studies have compared multiple predictive models, including linear regression, logistic
regression, and gradient boosting, to predict crime rates in urban areas. Gradient boosting has been
found to outperform other techniques effectively.
■ - Ensemble-based machine learning algorithms, such as random forest regressors, have been
applied to predict crime and quantify the impact of urban attributes on crime occurrences. These
models achieve high accuracy, with variables like unemployment and illiteracy identified as
significant predictors.
2) Machine Learning Based Classification Methods for Crime Prediction:
■ - Classification techniques have been explored for predicting different criminal incidents,
offering more reliable results for crime prediction. Novel approaches like incremental
classifiers and algorithms such as XGBoost have been utilized to dynamically classify and
predict crime reports.
■ - Techniques like SHAP (Shapley additive explanation) have been employed to determine the
significance of individual variables in crime prediction models. Variables such as the
percentage of the non-neighborhood population and the population aged 25–44 have been
identified as significant contributors to crime prediction.
■ - Machine learning classification techniques have also been applied to predict crimes during
or after psychiatric care, achieving significant accuracy in identifying psychiatric patients at
risk of committing crimes.
CRIME PREDICTION USING DEEP
LEARNING

■ Deep learning has emerged as a powerful method for crime prediction, leveraging
algorithms
■ like convolutional neural networks (CNN), deep neural networks, and sentiment analysis
across
■ various data types such as text, images, audio, and social media. These algorithms excel in
■ detecting patterns and anomalies in complex datasets, making them well-suited for crime
■ prediction tasks.

1) Deep Learning-Based Regression Methods for Crime Prediction:

■ - Deep learning regression techniques are employed to model the relationship between
crime data and various factors like weather, demographics, and social media data.
■ - Techniques like Long Short-Term Memory (LSTM) networks, Spatio-Temporal Graph
Convolutional Networks (ST-GCN), and convolutional neural networks are used to predict
theft crimes, urban crime trends, and crime occurrences in different cities.
■ - These regression models incorporate external information and capture temporal and spatial
dependencies in the data, resulting in improved accuracy and real-time performance.
ADVANTAGE
• Early Intervention: Crime prediction models can help law enforcement agencies intervene before
crimes occur, potentially preventing them or reducing their impact.
• Resource Optimization: By accurately predicting when and where crimes are likely to occur, law
enforcement agencies can allocate their resources more effectively, focusing on areas with higher
predicted crime rates.
• Data-driven decision-making: These models enable law enforcement to make decisions based on
data rather than intuition or past practices, potentially leading to more effective crime prevention
strategies.
• Identification of Patterns: Deep learning and machine learning algorithms can uncover complex
patterns in crime data that may not be apparent to human analysts, leading to insights that can
inform preventive measure
DISADVANTAGE
■ Bias in Data: Crime data often reflect existing biases in law enforcement practices, such as over-
policing of certain communities or under-reporting of certain types of crime. If not properly
addressed, these biases can be perpetuated and amplified by predictive models, leading to unfair
or discriminatory outcomes.
■ Privacy Concerns: Crime prediction models often rely on large amounts of data, including
personal information about individuals. There are concerns about the privacy implications of
collecting and analyzing such data, especially if it is used to target specific individuals or
communities.
■ Limited Predictive Accuracy: While these models can be effective in certain contexts, they are
not infallible. Predictive accuracy can vary depending on factors such as the quality of the data,
the complexity of the underlying patterns, and the assumptions made by the algorithm.
■ Ethical Implications: There are ethical considerations surrounding the use of predictive policing,
including concerns about fairness, transparency, and accountability. Deploying these models
without careful consideration of these issues can lead to unintended consequences and undermine
public trust in law enforcement.
APPLICATIONS
 Crime Type Prediction: These models can predict the type of crime that is likely to occur in a
given location and time, allowing law enforcement to tailor their response strategies accordingly.
For example, different tactics may be needed to address property crime compared to violent
crime.
 Trend Analysis: Crime prediction models can analyze historical data to identify trends and
patterns in criminal activity, helping law enforcement anticipate emerging threats and adapt their
strategies proactively.
 Risk Assessment: Predictive models can assess the risk of recidivism or future criminal behavior
among individuals in the criminal justice system, aiding in decisions related to pretrial release,
probation, and parole supervision.
CONCLUSION
The complexity of crimes has increased along with technological development, creating difficult problems for law
enforcement. Researchers’ interest in utilizing machine learning and deep learning to predict crime has increased
recently, with an emphasis on finding patterns and trends in crime occurrences. To analyze the various machine
learning and deep learning algorithms used in predicting crime, this paper looks at more than 150 articles. We have
significantly studied the selected 51 articles to extract the essence of utilized various ML and DL techniques along
with the publicly available datasets. The use of machine learning and deep learning algorithms to anticipate or
identify criminal activity has shown significant promise in resolving the crime detection problem. These advances
may help to increase the precision and efficacy of crime prediction models by leveraging large datasets and
sophisticated algorithms. Although there is a lack of literary wisdom on how these technologies can be used to
solve the problem of crime prediction, despite the advancements in this sector. Thus our findings help to understand
the implications of various ML and DL techniques. Also, our mentioned datasets and future directions will help the
existing research community to pursue their research in the area of crime prediction.
THANK YOU