Prediction Web Application Based on a Machine Learning Model to

Reduce Robberies and Thefts Rate in Los Olivos,

San Martín De Porres and Comas

Mederos Sanchez, Luis Estefano

, Zelada Padilla, Carlos Antonio

and Pedro S. Castañeda

Faculty of Information Systems Engineering, Universidad Peruana de Ciencias Aplicadas, Surco, Lima, Peru

Keywords: Machine Learning, Robbery, Thefts, Python, Random Forest Regressor, Web Application,

IBM Watson Learning Machine.

Abstract: Robberies and thefts in the districts of Los Olivos, San Martin de Porres and Comas in Lima, Peru are a

constant problem. The scarce police presence on the streets makes these areas ripe for crime. This project

proposes analyze crime rates across the public authorities to take measures that might reduce the crime rate

with the development of a Machine Learning model, through the use of Random Forest (RF) and a dataset

with information from districts in similar situations to those raised in the project. The proposed solution

includes a web application interface for data input and analysis, that will be used by municipal entities and

everyone. Performance metrics such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE)

were included, with results showing MAEs of 29.194, 45.219, and 75.572 and RMSEs of 39.651, 58.199, and

93.110 from other districts with the same condition. The study concludes with a refinement of machine

learning methodologies for crime prediction and emphasizes the potential for citizen engagement in crime

prevention.

1 INTRODUCTION

The issue of citizen insecurity is not only evident in

Lima's districts but also constitutes a nationwide

phenomenon of concern. As per an analysis

conducted in 2023 by the National Institute of

Statistics and Informatics (INEI), during the period

from November 2021 to April 2022, 21.1% of the

urban population in Lima aged 15 and older fell

victim to some form of criminal activity. This statistic

experienced a surge to 25% within the November

2023 timeframe.

Hence, it is of paramount importance to devise

strategic solutions that can assist Peru in reducing this

crime rate. It must be emphasized that attempting to

encompass all of Lima and every existing criminal

offense is not a viable option, as in terms of Machine

Learning, the indiscriminate use of extensive data in

an initial version can compromise the model's

accuracy. Therefore, the primary focus of this

https://orcid.org/0009-0008-5039-194X

https://orcid.org/0009-0009-5065-1411

https://orcid.org/0000-0003-1865-1293

solution will be on specific districts and types of

crimes.

In this context, the districts under consideration

are Los Olivos, San Martin de Porres (SMP), and

Comas. Furthermore, the targeted crimes will be

robberies and thefts. The rationale behind selecting

these focal points is their high crime rates, making

them known as crime hotspots. For instance, a 2022

ranking by the Legal Defense Institute (IDL)

identified Los Olivos, SMP, and Comas as the

districts with the highest volumes of robberies and

thefts, with estimated figures of 2232 and 2381; 1153

and 1192; 1650 and 2450, respectively.

Therefore, the objective of this project is to

provide citizens, governmental entities, and law

enforcement agencies with the opportunity to gain

greater insight into future crime rates in these

districts. This initiative enables entities to conduct

analyses and assessments, leading to more accurate

decision-making regarding crime, the reduction of

Sanchez, M., Estefano, L., Padilla, Z., Antonio, C. and Castañeda, P.

Prediction Web Application Based on a Machine Learning Model to Reduce Robberies and Thefts Rate in Los Olivos, San Martín De Porres and Comas.

DOI: 10.5220/0012906800003825

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 20th International Conference on Web Information Systems and Technologies (WEBIST 2024), pages 191-198

ISBN: 978-989-758-718-4; ISSN: 2184-3252

191

criminality, and the enhancement of public security

systems to uphold the well-being of the populace.

The system operates through a web application

where entities can input crime data. This information

flows into a database connected to a machine learning

model utilizing algorithms such as RF, producing

actionable insights. The following proposal is a

solution that, utilizing machine learning models,

enables predictions about the impact or priority of

crimes based on historical data. The subsequent

sections of this document are as follows: Section 2

conducts a comprehensive literature review of

various crime management solutions. In Section 3, a

meticulous depiction of the proposal is furnished,

accentuating the architecture, the proposed model and

the interface web solution. Section 4 elucidates the

findings, while Section 5 presents the conclusions and

future work.

2 BACKGROUND & RELATED

WORK

In 2020, a study explored the determinants of

organized crime by foreign terrorist criminals using

Machine Learning (ML) tools. They measured

environmental and organizational characteristics

within known terrorist organizations and applied

inductive research designs to examine criminal

behavior patterns, RF classification algorithms were

employed to predict when a terrorist organization

would engage in future criminal activities. Results

showed organizational factors outweighed

environmental factors in the classifier specifications,

suggesting organizational variables are crucial in

explaining the connection between terrorism and

organized crime (Semmelbeck and Besaw, 2020).

This study provided insights into the importance of

organizational variables and the efficacy of RF in

analyzing data patterns.

In a different context, another study aimed to

construct a predictive model to analyze crime data in

South Africa from 2005 to 2016. The goal was to

detect hidden patterns and generate reports for

implementing crime prevention measures within the

country. The research used linear regression

methodology to investigate crime trends, aiding

authorities. They utilized Python (PY) libraries for

data visualization to showcase correlations between

crimes in provinces over the established years. The

data parameters included registered crimes,

population density of each province, and the number

of police stations. The study found that the linear

regression model predicted crime rates in South

Africa with an accuracy of 84.7% indicating a strong

relationship between crime occurrence, population,

and province density (Obagbuwa and Abidoye,

2021).

Salcedo-González et al. (2023) focuses on

developing predictive geovisualization tools aimed at

controlling and preventing criminal activities, using

information provided by the National Police of

Colombia (PONAL). The study includes real-time

events for constant evaluation and training and the

used variables from the dataset: "timestamp,"

"latitude," "longitude," and "case code". Finally, the

results of the model with a better performance were

the 1D Convolutional Neural Network, with an

RMSE value of 0.285, indicating it is closer to zero.

Another study focused on developing a machine

learning algorithm to analyze criminals' anti-

investigation behavior, aiming to understand its

relationship with increasing crime rates. The Support

Vector Machine (SVM) algorithm was employed as

an effective classifier, utilizing decision boundaries to

separate data into two categories, with the goal of

identifying intelligent patterns allowing criminals to

commit crimes undetected through subsequent

investigations. Results concluded that, compared to

various algorithms, the RF algorithm achieved the

highest prediction accuracy, reaching 98.05% with 12

included features and 4770 support vectors (Zhang

and Lei, 2022). The experimental part highlights the

significant influence of criminals evading traditional

investigations on society. This approach would

confirm that the results truly impact problem

resolution.

In 2023, a study aimed to determine the most

effective machine learning models for predicting

criminal recidivism among convicted offenders in

Ukraine. Using artificial intelligence algorithms and

blockchain tools, the research analyzed a database

containing records of over 13,000 Ukrainian offenders

to identify factors influencing repeated offenses.

Various machine learning models such as DT, RF, and

SVM were employed to determine the most accurate

model. The study revealed that Gradient Boosted

Trees, RF, and DT techniques achieved prediction

accuracy levels of 98.3%. Furthermore, factors

contributing to criminal recidivism were identified,

including a 67% likelihood of reoffending among

convicts receiving lenient sentences, as well as

significant impacts of educational attainment,

particularly among first or second-time offenders.

Lastly, the number of suspended or actual convictions

played a role in the recidivism trend for third-time

offenders (Kovalchuk et al., 2023).

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

192

In the same way in 2023, a project for crime

prediction like homicides conducted in Bogota,

Colombia used data provided by the national police of

Colombia, covering the period from 2012 to 2017. In

the results, after training and testing the model, suggest

that the RF algorithm outperforms the other proposed

techniques, which are SVM, NBC, and KNN, proving

to be the most effective (Rodrigues et al., 2023).

Moreover, a study in Saudi Arabia aimed to

identify the most suitable ML algorithms for

predicting criminal activity in different regions.

Through various ML techniques, the Naive-Bayes

Classifier (NBC) was found to be the most accurate,

achieving a higher performance than other classifiers

in both Mixed Data Analysis (FAMD) and Principal

Component Analysis (PCA) methods, with an

accuracy of 97.53% and 97.10%, respectively

(Albahli et al., 2020). However, incorporating

different types of crimes in the model might have

impacted the results.

Additionally, a scientific article addressed the

prediction of potential targets for suicide attacks in

Pakistan using machine learning algorithms. By

analyzing terrorism data from the South Asia

Terrorism Portal (SATP), the study achieved notable

accuracy rates with algorithms such as Naive Bayes

(NB) and Sequential Minimal Optimization (SMO),

reaching precision rates of 72.17% and 71.30%,

respectively. These classifiers facilitated the

identification of specific individuals prone to

committing terrorist acts, including variables such as

location, day, month, province and city (Mahmood

and Ghani, 2021). This research shows a significant

advancement in using machine learning to predict

profiles of individuals engaging in terrorist activities.

In Roses et al. (2021), spatial crime simulation

techniques were developed to understand crime

mechanisms using machine learning on robbery data

from New York City between June 2014 and June

2015. Four simulation scenarios were created to

evaluate the model's performance. The first scenario

used the decision tree algorithm, with metrics like

RMSE and Predictive Accuracy Index (PAI). The

results indicate a PAI of 3% and an RMSE of 0.040.

Furthermore, research was conducted to produce

a prediction and forecasting model for crimes in

Chicago and Los Angeles using ML and Deep

Learning (DL) techniques. Eight different algorithms

were tested, including Support vector machine

(SVM), RF, DT, XGBoost, Multilayer Perceptron

(MLP), and logistic regression. The analysis yielded

two categories of results: accuracy percentages for

Chicago and Los Angeles data. XGBoost and logistic

regression algorithms achieved the highest accuracy

percentages for Chicago data, with 94% and 90%

accuracy, respectively. For Los Angeles data, the k-

nearest neighbors (KNN) algorithm demonstrated the

best performance, achieving an 89% accuracy, while

XGBoost reached an 88% accuracy (Safat et al.,

2021). It's important to note that predictions with

higher accuracy were obtained for crime data

collected in Chicago.

A study conducted an analysis to design an ML

model for predicting crimes in Porto, Portugal, by

combining data mining, geospatial technology, and

ML to identify high-risk areas and make accurate

predictions. Comparing various ML algorithms with

crime data from 2016 to 2018, the RF algorithm

yielded the best results, with a 99% True Positive

Rate (TPR) in testing and an 83% accuracy (Saraiva

et al., 2022). This research underscores the

importance of using logistic regression and RF in

similar projects due to their successful outcomes with

recent crime data.

In Baek et al (2021), an intelligent security system

is proposed to predict various types of crimes based on

a ML model, utilizing crime report summaries for

prediction. They primarily employ Deep Neural

Network (DNN) and Convolutional Neural Network

(CNN) architectures. Data is gathered from The Korea

Information System of Criminal Justice (KICS). The

DNN and CNN models are trained with a split of 60%

training (3000 crime cases), 20% validation (1000

crime cases), and 20% testing (1000 crime cases).

Metrics such as Accuracy, Precision, Recall, and F1-

Score are used to assess model accuracy. The CNN

model achieves the best results, with 91%, 92%, 82%,

and 84% in the mentioned metrics, respectively. The

authors' proposal represents a significant project in the

targeted area of study, as a real-time prediction model

will greatly assist law enforcement entities in decision-

making, whether it involves increasing police presence

in certain areas or taking specific actions in particular

cases.

In another research analysis, the city of Boston

has witnessed a significant shift in crime rates to the

extent that conducting a comprehensive and

analytical data analysis for better investigation

becomes challenging. The study aims to employ a

supervised learning approach by adding DT and RF

algorithms, along with Principal Component Analysis

(PCA), to the already mentioned algorithms. For the

study's development, they utilized records from the

Boston Police Department (BPD). The results

indicated that RF PCA achieved an accuracy of 60%

and an F1-score of 56%, followed by decision tree

PCA with an accuracy of 56% and an F1-score of

50% (Sharma et al., 2021). From the study, it can be

Prediction Web Application Based on a Machine Learning Model to Reduce Robberies and Thefts Rate in Los Olivos, San Martín De Porres

and Comas

193

highlighted that the precision levels are not high,

which is attributed to the poor distribution of crime

data, with few attributes contributing to better

prediction correlation.

Finally, in a study conducted in 2021, the focus

was on addressing the problem of crime using ML

models like RF algorithm to long-term forecast the

number of thefts in micro-locations in Dallas, Texas.

Additionally, performance was compared with other

techniques such as Risk Terrain Modeling (RTM) and

Kernel Density Estimation (KDE). The dataset used

was from the Dallas Open Data portal, where training

spanned from June 2014 to May 2016, with testing

from June 2016 to May 2018. The results showed that

RF outperformed the other techniques, with a PAI

value of 330.05, the highest (Wheeler and Steenbeek,

2021).

In summary, the literature reveals notable

advancements in using ML techniques to predict

various aspects of criminal behavior. However, there

are also differences and limitations within the studies.

Common findings include the effectiveness of ML

algorithms like RF and Support Vector Machine

(SVM) in crime prediction tasks across diverse

contexts. For instance, RF was found to be highly

accurate in predicting terrorist activities and criminal

recidivism, while SVM showed promise in

identifying patterns of anti-investigation behavior

among offenders. On the other hand, while some

focus on specific crime types or geographical regions,

others explore broader trends in criminal activity.

These differences underscore the multifaceted nature

of crime prediction and the need for tailored

methodologies to address specific challenges.

3 SYSTEM DESIGN

3.1 Architecture

The logical architecture of this project is structured to

accommodate different types of users, each with

specific roles and capabilities within the web

application. The system is initiated by either a user or

an administrator who accesses the application

through a web browser. Depending on their role, they

are presented with distinct interfaces and options.

Additionally, there is a guest user with limited

functionality, primarily granted access to some of the

options available to a regular user, such as obtaining

a prediction.

Both the administrator and the user have the ability

to request predictions. When a user submits a

prediction request, the web application communicates

with the model via Flask, filtering the uploaded

information to generate a prediction using the Random

Forest (RF) algorithm. During this process, the model

consults the database stored in MongoDB, where it

validates key variables such as date, type of robbery,

motorcycle use, amount, and weapon use, ensuring that

the data aligns with the necessary criteria for accurate

prediction. The administrator, beyond requesting

predictions, has the capability to upload new data,

which is crucial for enhancing the model's precision

over time. The administrator is also responsible for

overseeing the registration of users within the

application, ensuring effective management and

control.

Furthermore, the user can register a crime and

access valuable information on how to avoid

becoming a victim of theft or robbery. The detailed

functionalities and interactions of these components

are illustrated in the Figure 1.

3.2 Model

The model is based on RF algorithm, developed with

Python in Google Colab Platform. To train the model,

the dataset had to go through a data quality and

cleaning process. This phase involves checking

duplicate data, delete null data, and filter the data to

have just the information that the model requires.

Initially, the dataset counted with 169,000 registers,

after the data and quality and cleaning process the

dataset counts with 100,000 registers.

During the training phase, an additional column

was added to the dataset, the “Pandemic” field. This

variable is important for the training process because,

before the COVID-19 pandemic situation, the crime

rate was high. However, after the pandemic began,

the crime rate decreased a lot, because most people

were at home.

This variable helped the model to identify the

pandemic situation and give a better prediction. The

information used for the training phase was thefts and

robberies registered from 2016 to 2018, and the

information used to validate the prediction is from

July 2021 to December 2022.

Finally, three more columns were considered: the

first was "previous month", the second "two months

ago", and the last "three months ago". The reason for

this is that being a regression column of the total

number of crimes recorded, including those recorded

consecutively over the past three months.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

194

Figure 1: Logical Architecture of the Web Application.

3.3 Interface

The web application is specifically designed for

police or government entities, using technologies

such as ReactJS, HTML5 and CSS to provide an

intuitive and responsive interface. This platform

allows for the input and reception of data, which are

stored in a database and analyzed using the IBM

Watson machine learning model implemented in

Python. The analysis results are presented

quantitatively and graphically, facilitating their

understanding and use for informed decision-making.

Additionally, the interface allows for constant

adjustments and feedback, optimizing the accuracy of

the results. The user accesses into user interface (UI)

to appreciate all the functionalities offered by the web

frontend, while the web server backend handles client

requests and communicates with the Machine

Learning API. This API provides the interface to

interact with the Machine Learning model, including

the trained models, data processing, and the services

it offers. Additionally, security is a fundamental

feature in preventing potential threats to data loss.

Therefore, a security and authentication layer has

been implemented to protect the Machine Learning

model and input data, ensuring the integrity and

confidentiality of the information

Finally, it is important to highlight that the web

application offers different functionalities based on the

user type. The main features available for each group

of users within the interface are presented below:

Municipal Entity:

• Update Dataset: Allows for the uploading and

updating of relevant datasets for analysis.

• Generate Reports: Facilitates the creation of

detailed reports based on the collected data.

• Verify Users: Enables the management and

verification of registered users.

Residents of Comas, SMP or Los Olivos:

• User Registration: Simplifies the registration

process for new users.

• Crime Registration: Registration of crimes by

registered users.

• View Charts: Provides access to interactive

charts for a better understanding of relevant

information.

Prediction Web Application Based on a Machine Learning Model to Reduce Robberies and Thefts Rate in Los Olivos, San Martín De Porres

and Comas

195

Figure 2: Crime Reporting Interface.

In figure 2, this section of the web application

represents the interface for registering crimes by users

or municipal entities. They can record information

about the complaint such as the date, type of crime,

frequency of occurrence of this crime, district it

belongs to, and lastly, characteristics such as whether

a weapon was used or if a motorcycle was involved,

this information will be utilized in the future to assess

the model's performance and support the authorities

by contributing to the collection of this information.

3.4 Dataset

A specific dataset has been selected for the proposed

solution and the development of the ML model. This

dataset focuses on the districts in Argentina and is

sourced from the Buenos Aires (BA) Data platform,

which aggregates over 300 datasets from 13

government areas (Ministerio de Seguridad de la

Nación, 2023). The change of information is based on

the dataset's robust and comprehensive structure,

evident in the columns, compared to what was found

in research platforms related to Lima Metropolitana.

The following table presents the dataset design:

Table 1: Dataset from BA data.

Date

Type Weapon Motor

cycle

District

Pandem

10/14/

2016

Robbery NO NO Balvanera NO

05/23/

2022

Theft NO NO Villa

ano

YES

This dataset has many key columns with vital

information. It first contains a column of ‘Date’

which is a time series data of each event, this column

provides the analysis with temporal context to

identify patterns and trends in relation to the nature of

crimes.

The ‘Type’ column shows the nature of the event

including for example robbery and theft, providing

context for the type of criminal activity. The

‘Weapon’ column specifies whether a weapon was

involved in the crime, indicated as "YES" or "NO",

same case with the column ‘Motorcycle’. The column

of ‘District’ identifies the exact place where the event

took place, allowing the analysis of crime

distribution.

Finally, the column ‘Pandemic’ is an indication of

the crime occurred during the COVID-19 pandemic

or did not, considering the date between January 2020

and the end of 2022.

3.5 Indicators

The performance evaluation of the solution requires a

deep understanding of several key indicators. These

indicators provide a comprehensive view of the

system’s success and effectiveness in crime

prevention.

The MAE is a metric used to measure the

difference between two values and it indicates how

different the predicted value is from the actual or

observed value. For example, if we have actual

numbers like 3, 2.5 and the following predictions are

3.1, 2.5 so this indicates that the MAE value of 0.249

is the magnitude of errors in the observations, which

is satisfactory because the error is closer to zero

(Landa, 2021).

Finally, to understand the metric RMSE. First the

MSE was defined, which measures the average

squared error of the predictions, because calculates

the squared difference between the predictions and

the target. A higher MSE value indicates a worse

model and it is always non-negative. RMSE is the

square root of MSE, this square root adjustment

ensures that the error scale matches the scale of the

target values (Big Data, 2018).

4 RESULTS & DISCUSSION

Finally, in the results phase, the model demonstrated

robust performance across the tests in the districts in

Buenos Aires with the same crime situation as

districts in Lima, Peru, so the results obtained can be

implemented in the data structure in Lima to be able

to generate predictions in the future with the objective

of being a support to combat crime. The principal

metrics used to evaluate the model's accuracy were

RMSE and MAE. The model showed the following

results: in Villa Lugano, the MAE was 29.194 and the

RMSE was 39.651; in Recoleta, the MAE was 45.219

and the RMSE was 58.199; and in Balvanera, the

MAE was 75.572 and the RMSE was 93.110.

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

196

Figure 3: Bar chart comparing actual data and model predicted data in the Villa Lugano neighbourhood from 2021 to 2022.

In Villa Lugano, the lower MAE and RMSE

indicate that the model's predictions were very close

to the actual data, reflecting high accuracy. Recoleta,

while showing slightly higher error metrics, still

demonstrated reasonable predictive performance. In

Balvanera, the model's higher MAE and RMSE

suggest more significant deviations from the actual

values, indicating limitations, particularly when

confronted with extensive datasets with a higher

incidence of reported events.

During the prediction process, the model

consulted the MongoDB database to validate key

variables such as date, type of robbery, motorcycle

use, amount, and weapon use. This step ensured that

the data used for predictions was accurate and aligned

with the criteria necessary for reliable results.

However, despite these validations, the larger and

more complex datasets, particularly in areas like

Balvanera, presented challenges that resulted in

higher error margins compared to other districts.

As depicted in Figure 3, the model demonstrates

strong performance, particularly in Villa Lugano.

Instances of accurate forecasting include August

2021, where both real data and predictions align at

5.67%, December 2021 with a minimal deviation of

5.72% versus 5.75%, July 2022 showing a close

match at 5.55% and 5.61%, and finally, August 2022

displaying 5.65% against 5.61%. Additionally, it has

been found that the data in the columns, as well as the

context of the pandemic, have significantly affected

the performance of this type of this type of model.

Figure 4 shows that excluding factors like

consecutive months and the pandemic scenario, and

only considering the month and year, significantly

worsened the model's performance, with an RMSE of

39.46 and a MAE of 33.43, often deviating from real

data. The validation against MongoDB emphasized

the importance of including comprehensive data, such

as date, robbery type, motorcycle use, amount, and

weapon use, to improve prediction accuracy.

In summary, the metrics RMSE and MAE were

primary in this accuracy assessment, emphasizing the

necessity of discovering and incorporating more

detailed data to enhance predictive accuracy.

Figure 4: Line Chart from Villa Lugano (2021-2022).

5 CONCLUSION AND FUTURE

WORK

According to this study, the ML model demonstrates a

strong ability to generate precise predictions, closely

aligning actual and predicted criminal rates, especially

when considering the pandemic's mitigating impact on

crime. This means that using the model with real-time

data will yield accurate results, effectively helping the

project achieve its main objective. Unlike previous

Prediction Web Application Based on a Machine Learning Model to Reduce Robberies and Thefts Rate in Los Olivos, San Martín De Porres

and Comas

197

related works, this project offers significant added

value through the integration of a web application and

continuous data updates, ensuring that predictions are

based on the most current information available. The

project addresses the pressing issue of rising crime in

specific districts and crime types, providing valuable

insights for decision-making to enhance public security

systems and reduce criminality. It also promotes

citizen engagement through the web application and

user interfaces.

This research enhances crime prediction by

developing an RF-based model that considers key

factors affecting accuracy. To further improve results,

the study suggests integrating advanced technology

and refining strategies to meet stakeholder needs.

ACKNOWLEDGEMENTS

The authors are grateful to the Dirección de

Investigación de la Universidad Peruana de Ciencias

Aplicadas (UPC) for the support provided for this

research work through the incentive.

REFERENCES

Albahli, S., Alsaqabi, A., Aldhubayi, F., Rauf, H. T., Arif,

M. and Mohammed, M. A. (2020). Predicting the type

of crime: Intelligence gathering and crime analysis.

Computers, Materials & Continua 2021, 66(3), 2317-

2341. https://doi.org/10.32604/cmc.2021.014113

Baek, M. S., Park, W., Park, J., Jang, K. H., and Lee, Y. T.

(2021). Smart Policing Technique with Crime Type and

Risk Score Prediction Based on Machine Learning for

Early Awareness of Risk Situation. IEEE Access,

9(2021), 131906-131915. https://doi.org/10.1109/

ACCESS.2021.3112682

Big Data. (2018). Aprendizaje Automático ML: Métricas de

Regresión. https://sitiobigdata.com/2018/08/27/machi

ne-learning-metricas-regresion-mse/

INEI. (2023). Estadísticas de Seguridad Ciudadana.

https://m.inei.gob.pe/media/MenuRecursivo/boletines/

estadisticas-de-seguridad-ciudadana-noviembre-2022-

abril-2023.pdf

Kovalchuk, O., Karpinski, M., Banakh, S., Kasianchuk, M.,

Shevchuk, R., and Zagorodna, N. (2023). Prediction

Machine Learning Models on Propensity Convicts to

Criminal Recidivism. Information (Switzerland),

(2023), 14(3). https://doi.org/10.3390/info14030161

Landa, N, (2021). Métricas en Regresión. Medium.

Mahmood, N and Ghani, M. (2021). Prediction of Extremist

Behaviour and Suicide Bombing from Terrorism

Contents Using Supervised Learning. Computers,

Materials and Continua (2022), 70(3), 4411-4428.

https://doi.org/10.32604/cmc.2022.013956

Ministerio de Seguridad de la Nación. (2023). SNIC -

Provincial. Estadísticas criminales en la República

Argentina por provincias. https://datos.gob.ar/dataset/

seguridad-snic---provincial-estadisticas-criminales-

republica-argentina-por-provincias

Obagbuwa, I., and Abidoye, A. (2021). South Africa Crime

Visualization, Trends Analysis, and Prediction Using

Machine Learning Linear Regression Technique.

Applied Computational Intelligence and Soft Computing,

(2021). https://doi.org/10.1155/2021/5537902

Rodrigues, A., González, J. A., and Mateu, J. (2023). A

conditional machine learning classification approach

for spatio-temporal risk assessment of crime data.

Stochastic Environmental Research and Risk

Assessment, (2023). https://doi.org/10.1007/s00477-

023-02420-5

Roses, R., Kadar, C., and Malleson, N. (2021). A data-

driven agent-based simulation to predict crime patterns

in an urban environment. Computers, Environment and

Urban Systems, 89(2021). https://doi.org/10.1016/j.

compenvurbsys.2021.101660

Salcedo-Gonzalez, M., Suarez-Paez, J., Esteve, M., and

Palau, C. (2023). Spatiotemporal Predictive Geo-

Visualization of Criminal Activity for Application to

Real-Time Systems for Crime Deterrence, Prevention

and Control. ISPRS International Journal of Geo-

Information, (2023), 12(7). https://www.mdpi.com/

2220-9964/12/7/291

Safat, W., Asghar, S., and Gillani, S. A. (2021). Empirical

Analysis for Crime Prediction and Forecasting Using

Machine Learning and Deep Learning Techniques.

IEEE Access, 9(2021), 70080-70094. https://doi.

org/10.1109/ACCESS.2021.3078117

Saraiva, M., Matijošaitienė, I., Mishra, S., and Amante, A.

(2022). Crime Prediction and Monitoring in Porto,

Portugal, Using Machine Learning, Spatial and Text

Analytics. ISPRS International Journal of Geo-

Information, (2022), 11(7). https://doi.org/10.3390/

ijgi11070400

Semmelbeck, J. and Besaw, C. (2020). Exploring the

Determinants of Crime-Terror Cooperation using

Machine Learning. Journal of Quantitative Criminology,

(2020), 36(3), 527-558. https://doi.org/10.1007/s10940-

019-09421-0

Sharma, H., Choudhury, T., and Kandwal, A. (2021).

Machine learning based analytical approach for

geographical analysis and prediction of Boston City

crime using geospatial dataset. GeoJournal, (2021).

https://doi.org/10.1007/s10708-021-10485-4

Wheeler, A., and Steenbeek, W. (2021). Mapping the Risk

Terrain for Crime Using Machine Learning. Journal of

Quantitative Criminology, (2021), 37(2), 445 - 480.

https://doi.org/10.1007/s10940-020-09457-7

Zhang, J and Lei, Y. (2022). Trend and Identification

Analysis of Anti-investigation Behavior in Crime by

Machine Learning Fusion Algorithm. Wireless

Communications and Mobile Computing, (2022).

https://doi.org/10.1155/2022/1761154

WEBIST 2024 - 20th International Conference on Web Information Systems and Technologies

198