Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Antisource/FFD/llms.txt

Use this file to discover all available pages before exploring further.

The FFD project is built on a publicly available credit card transaction dataset hosted on Kaggle. It contains 50,492 real-world transactions made by European cardholders, each labelled as either genuine or fraudulent. The dataset is widely used in the machine learning community as a benchmark for fraud detection, and it presents a classic extreme class imbalance challenge that the FFD pipeline directly addresses through GAN-based data synthesis.

​
Dataset Structure

The dataset has 50,492 rows and 31 columns. Each row represents one transaction, and the columns are as follows:
ColumnTypeDescription
TimeFloatSeconds elapsed between this transaction and the first transaction in the dataset
V1 – V28FloatPCA-transformed anonymized features (original names are confidential)
AmountFloatTransaction amount in euros
ClassIntegerTarget label: 0 = genuine transaction, 1 = fraudulent transaction

​
Shape

data.shape
# (50492, 31)

​
Class Distribution

data.Class.value_counts()
# 0    50000
# 1      492
# Name: Class, dtype: int64
The 28 PCA-transformed features (V1–V28) are the result of a dimensionality reduction applied to the original transaction attributes, which remain confidential for privacy reasons. Only Time and Amount retain their original, interpretable values.

​
Loading the Dataset

import pandas as pd

data = pd.read_csv('Creditcard_dataset.csv')

​
Class Imbalance at a Glance

The dataset contains 50,000 genuine transactions (Class = 0) and only 492 fraudulent transactions (Class = 1), making fraud just ~0.975% of all records. This severe imbalance means a naive model can achieve over 99% accuracy by simply predicting every transaction as genuine — while completely failing to detect any actual fraud.

Genuine Transactions

50,000 labelled Class = 0

Fraudulent Transactions

492 labelled Class = 1
The FFD pipeline resolves this imbalance by training a Generative Adversarial Network (GAN) to synthesize 492 additional realistic fraud samples, bringing the training ratio closer to 1:1. See Class Imbalance for the full explanation.

​
Accessing the Dataset

1

Download from Kaggle

Visit the Credit Card Fraud Detection dataset page on Kaggle and download creditcard.csv. A free Kaggle account is required.
2

Rename and place in the repo root

Rename the downloaded file to Creditcard_dataset.csv and place it in the root of the cloned FFD repository alongside Design_Project.ipynb.
3

Verify the load

Open the notebook and run the data-loading cell to confirm data.shape returns (50492, 31).

Build docs developers (and LLMs) love

Get started for freeTalk to us