I am working in a Synthetic Database for Fraud Detection in banking transactions. This is another Kaggle project with over 6 million rows, being each row a bank transaction. One of the challenges is that this is a very unbalanced database with a very small amount of fraudulent transactions (less than 1%). Let’s start working.

Brief description of Fraud Detection.

1. Dataset contains 6,362,620 transactionsm from a synthetic dataset generated using the simulator called PaySim. “The simulator uses aggregated data from the private dataset to generate a synthetic dataset that resembles the normal operation of transactions and injects malicious behaviour to later evaluate the performance of fraud detection methods”, as explained in Kaggle.
2. The baseline is 99.87% of non-fraudulent mobile money transactions.
3. The link in github for a more detailed and technical explanation is not yet available.

Non-technical description of Appointments No-show study.

Gathering the data.

The file happens to be pretty clean with no nulls but some inconsistencies as I will point out.

First one:

1. THe field ‘isFlaggedFraud” is supposed to be the transactionsDataset contains 6,362,620 transactionsm from a synthetic dataset generated using the simulator called PaySim. “The simulator uses aggregated data from the private dataset to generate a synthetic dataset that resembles the normal operation of transactions and injects malicious behaviour to later evaluate the performance of fraud detection methods”, as explained in Kaggle.
2. The baseline is 99.87% of non-fraudulent mobile money transactions.
3. The link in github for a more detailed and technical explanation is not yet available.

python

file = 'PS_20174392719_1491204439457_log.csv'
df = pd.read_csv(file)

data = df.copy()

data = data.rename(columns={'oldbalanceOrg': 'oldbalanceOrig'})

data.shape