Credit card fraud are ubiquitous in nature. Every company in today's world is employing machine learning to combat payment fraud on their platform. In this chapter, we looked at the problem of classifying fraud using the credit card dataset from Kaggle.
We learned about auto-encoders as a dimensionality reduction technique. We understood that the auto-encoder architecture consists of two components: an encoder and a decoder. We model the parameters of a fully connected network using reconstruction loss.
Thereafter, we looked at the fraud classification problem through the lens of an anomaly detection problem. We trained the auto-encoder model using normal transactions. We then looked at the reconstruction error of the auto-encoder for both normal and fraudulent transactions, and observed that the reconstruction error has a wide distribution for fraudulent transactions. We then defined a threshold on reconstruction to classify the model and generated the confusion matrix.
In the next...