A Transformer-based Model Integrated with Feature Selection for Credit Card Fraud Detection
Yuan Miao · 2022
Credit card fraud has grabbed a lot of attention in data science domain in the past several years. Most previous works tackled this problem by using machine learning methods equipped with hand-crafted features. In this paper, we proposed a novel multi-stage method combing both feature selection by a machine learning method, as well as a transformer-based model as a classifier. Concretely, we firstly extracted the top-k important features from the original dataset, where they are further used to train the self-designed deep learning model. Experimental results shown that our model can achieve better results compared to the baseline models. Since we also incorporated transformer layer into our deep learning model, we also tested its performance and the results proves the effectiveness.