Classificação do perfil dos clientes através de técnicas de mineração de dados

• Autor(a) principal: Santos, Matheus Henrique
• Data de Publicação: 2023
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/38729
• topic: Aprendizado de Máquina; Regressão Logística; Árvore de Decisão; Perfil de Clientes; Machine Learning; Logistic Regression; Decision Tree; Customer Profile; CNPQ::CIENCIAS EXATAS E DA TERRA::PROBABILIDADE E ESTATISTICA::ESTATISTICA

Currently, ever-increasing volumes of data are generated, but they are not always used in an “intelligent” way to extract information. In this scenario, data mining tools are used to help extract information that is not always so obvious or explicit, which makes it one of the most promising technologies today. The present work was developed with the purpose of creating a classification model, through data mining techniques, for the classification of the profile of the customers (Customer for Sale or Customer Not for Sale) of a real company. The structure of this Case Study used the CRISP-DM methodology, which is the most common in problems involving Data Mining. This methodology is segmented into 06 stages, which are: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation and Use. Two Classification algorithms were tested for model development: Decision Tree and Logistic Regression. For application in the algorithms, the database was divided into two bases: Training Base (composed of 70% of the data) and Test Base (Comprised of 30% of the data). As this is a database with unbalanced data, due to the significant superiority of customers in the Non-Sales class, two methodologies were evaluated for data balancing: application of SMOTE (Synthetic Minority Oversampling Technique) for oversampling technique and adjustment of weights of the minority and majority classes within the cost functions of the algorithms. The Logistic Regression models both for the application of SMOTE and for adjusting the weights of the classes presented similar results, and were the ones that presented the best results. With this, the study concluded that the classification model found, through the Logistic Regression technique, is extremely relevant for the company, resulting in a reduction in the costs of contacting customers and a significant increase in the volume of return of these contacts (Sales).