The problem Financial Fraud Detection solves

It plans on detecting financial frauds occuring by taking in transaction id from huge set of data and finding its column values which it sends to our backend ML model which then predicts if the transaction is Fraud, or not Fraud

Challenges we ran into

• We found that 3 types of transactions had 0 Fraud Cases so we decided to neglected them from training and created certain conditions to judge if their transactions are Fraud or not which are explained in our Data Exploration Notebook
• Dataset was completely unbalanced so we decided to randomly select our data based on 1:1 ration of Fraudulent and non-Fraudulent Data from our dataset which we trained our model on and got a 99% Recall and 99% F1-Score metrics on our data.
• We also had some deployment issues which we faced due to large dataset