- E-commerce has really changed everything, it gives us the chance to increase our sales but it exposes us to hackers and other types of frauds.
- For this case study we will consider a E-commerce book store which has sold thousands in the last few years.
- We are going to use transaction history for this project.
- We are going to use a publicly available dataset for this project with real credit-card transactions that have been anonymized.
- one of the biggest problem of credit card is that anyone can steal 15 or 16 digit card number with security number and expiration date.
- Our job will be to tell apart that the transaction was from the real user who owns the credit card or the transaction was from the stolen credit card.
- One of the major issues will be that most of the transactions will be non-fraudulent which means it will be difficult for us to detect underlying patterns in the infromation available.
- It will make our dataset highly imbalance.
- We will need to apply different sampling techniques and use different metrics.
- It is estimated that only 0.1% of online credit card transactions are fraudulent.
- But given the volume of transactions that occur everyday that means a lot of money.
- We need to build the classifier system based on the dataset that can tell wether the classifier is fraudulent or non-fraudulent.
- We will build a deep learning network and we are going to apply more traditional machine learning algorithm such as random forest.
- We will consider only 2 labels in our classifier '0' for non-fraudulent transaction and '1' for fraudulent transactions.