Fraud detection is a critical area where deep learning has been effectively applied to identify and prevent unauthorized transactions, money laundering, and other financial crimes. Traditional rule-based systems and statistical models often struggle to detect sophisticated fraud patterns, particularly when dealing with large volumes of data and rapidly evolving fraud techniques. In contrast, deep learning models, such as CNNs, RNNs, and autoencoders, have proven highly effective in analyzing complex, high-dimensional transaction data and detecting subtle, non-linear patterns indicative of fraudulent activity.
In this project, we build a User ID-based fraud detection model using autoencoders, trained on unlabelled real-world credit card transaction data, capable of detecting fraud with a precision of up to 35% and a recall of up to 72%, performing significantly better than traditional ML/statistical baseline models..

In this project, trained a cycleGAN on speech data and singing data to create a voice synthesizer that takes speech and outputs a synthesized voice to play over a given song.