Anomaly Detection for Fraud Prevention - University Project

In this project, it is proposed to explore the use of unsupervised machine learning
techniques able to address fraud detection. By applying this techniques we should be
able to model normality and detect any deviation from this normality as an anomaly or
potential fraud. We aim to evaluate this methodologies in fraud datasets such as
https://www.kaggle.com/mlg-ulb/creditcardfraud and https://www.kaggle.com/ntnu/testimon/banksim1

Objectives

Study current unsupervised machine learning techniques, particularly state-ofthe-art on fraud detection
• Investigate how transaction features cane be extracted to differentiate normal
and abnormal cases
• Investigate unsupervised machine learning techniques based on clustering and
autoencoder to model normal transaction.
• Implement an anomaly detection mechanism that using the previous model can
detect outliers as potential fraud cases
• Evaluate the performance of the proposed system and compare it against the
state of the art in the field using standard datasets and appropriate metrics

Skills
This project is best suited to a person with an interest in Deep learning and strong
skills in programming and mathematics