| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.92 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A utilização do comércio eletrónico, mais conhecido por E-Commerce, aumentou nos últimos anos, eventualmente devido à pandemia covid-19 bem como a outros fatores. Este aumento levou à criação de novas empresas, mas também estimulou diversas atividades de fraude online. As empresas cada vez mais têm dificuldades para combater e impedir tentativas de fraudes devido à sua constante mudança. Esta dissertação visa apresentar todo o processo de estudo, experimentação, desenvolvimento e avaliação de uma solução de Machine Learning para a deteção de fraudes, desde a recolha dos dados ao seu processamento e apresentação de resultados. Durante o desenvolvimento foram aplicados conhecimentos obtidos na fase de estudo e técnicas de Machine Learning. O objetivo é comparar os diferentes algoritmos [Decision Tree (DT), Isolation Forest (IF) e K-nearest neighbors (KNN)] e verificar quais oferecem melhor resultado a cada conjunto de dados[Clientes e Transferências], na deteção de fraude ou outliers. Um objetivo secundário, tirando partido do uso de Machine Learning, será efetuar previsões nomeadamente o nível de risco de um novo cliente por Merchant. Nesta tese pode ser encontrado um estudo e comparação da aplicação de dois modelos, um para a previsão de fraude de um cliente e outro para a deteção de outlier nas transferências, assim como um estudo sobre as melhores configurações e parâmetros para cada modelo.
The use of electronic commerce, commonly known as E-commerce, has increased in recent years, possibly due to the Covid-19 pandemic as well as other factors. This growth has led to the creation of new companies but has also stimulated various online fraud activities. Companies are increasingly facing challenges in combating and preventing fraud attempts due to their constantly evolving nature. This dissertation aims to present the entire process of studying, experimenting, developing and evaluating a Machine Learning solution for fraud detection, from data collection to processing and presentation of results. During the development, knowledge gained in the study phase and Machine Learning techniques were applied. The goal is to compare different algorithms [Decision Tree (DT), Isolation Forest (IF), and K-nearest neighbors (KNN)] and determine which ones offer the best results for each data set [Customers and Transfers] in fraud or outlier detection. A secondary objective, taking advantage of the use of Machine Learning, is to make predictions, particularly the risk level of a new customer by Merchant. This thesis includes a study and comparison of the application of two models, one for predicting customer fraud and another for outlier detection in transfers, as well as a study on the best configurations and parameters for each model.
The use of electronic commerce, commonly known as E-commerce, has increased in recent years, possibly due to the Covid-19 pandemic as well as other factors. This growth has led to the creation of new companies but has also stimulated various online fraud activities. Companies are increasingly facing challenges in combating and preventing fraud attempts due to their constantly evolving nature. This dissertation aims to present the entire process of studying, experimenting, developing and evaluating a Machine Learning solution for fraud detection, from data collection to processing and presentation of results. During the development, knowledge gained in the study phase and Machine Learning techniques were applied. The goal is to compare different algorithms [Decision Tree (DT), Isolation Forest (IF), and K-nearest neighbors (KNN)] and determine which ones offer the best results for each data set [Customers and Transfers] in fraud or outlier detection. A secondary objective, taking advantage of the use of Machine Learning, is to make predictions, particularly the risk level of a new customer by Merchant. This thesis includes a study and comparison of the application of two models, one for predicting customer fraud and another for outlier detection in transfers, as well as a study on the best configurations and parameters for each model.
Description
Keywords
Machine Learning E-Commerce Deteção de Fraude Deteção de Outlier Fraud Detection Outlier Detection