Banca de QUALIFICAÇÃO: RAMIRO DE VASCONCELOS DOS SANTOS JÚNIOR

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : RAMIRO DE VASCONCELOS DOS SANTOS JÚNIOR
DATE: 19/04/2023
TIME: 10:00
LOCAL: Sala do Google Meet - http://meet.google.com/ycm-tafa-cag
TITLE:

Implementation of Supervised and Unsupervised Machine Learning Models for the Classification of Criminal Macrocauses: An Approach to Assist in the Analysis of Violent Crimes in the Area of Public Safety

KEY WORDS:

Crime Analysis, Criminal Macrocause, Machine Learning, Predictive Policing, Public Safety, Smart Cities

PAGES: 78
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUBÁREA: Matemática da Computação
SPECIALTY: Modelos Analíticos e de Simulação
SUMMARY:

In the design of effective public safety policies, intelligence and information are essential factors. The classification of criminal macrocauses can help governments better understand aspects of the analysis of intentional lethal violent crimes in their states. However, in some locations, such as Rio Grande do Norte (RN), a Brazilian state, crime analysts analyze these crimes based on raw datasets. It is an approach known as traditional, in addition to being a time-consuming and imprecise process. Furthermore, two or more experts may have different opinions about the same violent crime. Thus, we propose a predictive model capable of classifying criminal macrocauses from a set of predefined types. This model aims to help criminal analysts by making the analysis and management process more efficient and accurate. We applied the Design Science methodology to execute the defined methods and developed a pre-processing of resources. Four classification techniques were analyzed: Decision Trees, Logistic Regression, Random Forest, and XGBoost. After conducting statistical tests, we concluded that the model with XGBoost reached an average accuracy of 0.961791, an F1-Score of 0.961410, and the ROC curve obtained was 0.994732, which, according to the literature, it is considered excellent for this application. Complementing supervised machine learning techniques, we propose an unsupervised approach based on clustering techniques to compose the range of solutions. These specific algorithms aim to allow the creation of clusters with the characteristics of the criminal data set, generating insights and guiding experts during decision-making. Therefore, the scientific contribution of this study lies in proposing a model based on machine learning techniques, which involve classification and clustering techniques for analyzing criminal macrocauses. Given these conditions, these models can contribute to more efficient decision-making by crime analysts and allow the development of more effective public safety policies. Finally, the proposal for a model integration platform in the context of smart cities offers a necessary, agile solution for analyzing criminal macrocauses.

COMMITTEE MEMBERS:
Presidente - 1678918 - NELIO ALESSANDRO AZEVEDO CACHO
Externo ao Programa - 2579664 - ALLAN DE MEDEIROS MARTINS - UFRNExterno ao Programa - 1669545 - DANIEL SABINO AMORIM DE ARAUJO - UFRNExterna ao Programa - 1362181 - ISMENIA BLAVATSKY DE MAGALHÃES - UFRNExterna à Instituição - THAIS GAUDENCIO DO REGO - UFPB