Optimización Bayesiana en modelos de clasificación: Árbol de Decisión y Support Vector Machine para determinar mediante Minería de Datos patrones en los asesinatos de la Zona 8 del Ecuador

This study focused on improving data mining models through techniques such as Bayesian Optimization, with the objective of finding the appropriate hyperparameters to configure the Decision Tree (DT) and Support Vector Machine (SVM) classifiers, in addition, it focused on determining patterns in the murders occurred in Zone 8 of Ecuador, following the CRISP-DM methodology, which includes the phases of understanding the business, understanding the data, preparing the dataset, modeling, evaluation and deployment. The SVM model, specifically configured with the Optuna library, obtained the best results, reaching 83.78% with the cross validation in the three key metrics that were precision, accuracy and recall, while when evaluated with sklearn.metrics, the model reached 84.12% for each of these metrics; this model allowed identifying significant patterns, such as the firearm is the most used in the murders, these crimes occur mainly in urban areas mostly on public roads, on Saturdays and Sundays between 19:00 pm and 00:59 am, especially in the Districts of Nueva Prosperina, Sur and Pascuales, as for the victims of the murders are usually men who do have criminal records, in ages ranging between 20 and 50 years. With these patterns found, valuable information was provided to the agencies in charge, which contributed to a better understanding of the phenomenon studied. Furthermore, this study evidenced that Bayesian Optimization considerably improves classification models by adjusting their hyperparameters, increasing performance percentages, such as precision, accuracy, sensitivity, and thus strengthening their practical applicability.

• Hoofdauteur: Trueba Reyes, Cecilia Fernanda (author)
• Formaat: bachelorThesis
• Taal: spa
• Gepubliceerd in: 2025
• Onderwerpen: CONFIGURACION; MINERÍA DE DATOS; ALGORITMOS
• Online toegang: https://dspace.unl.edu.ec/jspui/handle/123456789/32090