Fraud Detection Model

This repository contains a PyTorch implementation of a neural network for credit card fraud detection.

Accuracy: 0.9994

Precision: 0.8144

Recall: 0.8061

F1-score: 0.8103

Overview

The notebook walks through the process of building and training a binary classification model using a dataset containing anonymized credit card transactions. The goal is to identify fraudulent transactions.

Dataset

The model is trained on a dataset that includes transaction details such as time, amount, and 28 anonymized features (V1-V28), along with a 'Class' label indicating whether the transaction is fraudulent (1) or not (0).

Model Architecture

The model is a simple feedforward neural network with two hidden layers and dropout for regularization.

• Input Layer: Matches the number of features in the dataset.
• Hidden Layer 1: 64 neurons, followed by ReLU activation and dropout.
• Hidden Layer 2: 32 neurons, followed by ReLU activation and dropout.
• Output Layer: 1 neuron with a linear activation (sigmoid is applied during evaluation for probability).

Dependencies

The following libraries are required to run the notebook:

• torch
• torch.nn
• torch.nn.functional
• sklearn.model_selection
• sklearn.preprocessing
• torch.optim
• numpy
• pandas
• sklearn.metrics

Usage

1. Load the data: Load the creditcard.csv file into a pandas DataFrame.
2. Preprocessing:
   • Separate features (X) and labels (y).
   • Convert data to PyTorch tensors.
   • Handle any potential NaN values.
   • Split the data into training and testing sets.
   • Scale the features using StandardScaler.
3. Model Definition: The FraudDetectionModel class defines the neural network architecture.
4. Training:
   • Instantiate the model, loss function (BCEWithLogitsLoss with pos_weight to handle class imbalance), and optimizer (Adam).
   • Create DataLoader instances for the training and testing datasets.
   • Train the model using the train_model method.
5. Evaluation: Evaluate the trained model on the test set using the evaluate_model method, which calculates accuracy, precision, recall, and F1-score. A threshold is applied to the model's output to classify predictions.

Files

• creditcard.csv: The dataset containing the transaction data.
• This notebook: Contains the Python code for the model implementation, training, and evaluation.

🔗 Download the dataset

Results

The evaluation metrics (Accuracy, Precision, Recall, F1-score) on the test set are printed after the model is evaluated.