IEEE Fraud Detection Using AutoGluon

Project Overview

This project aims to build a robust machine learning model to detect fraudulent transactions using the IEEE-CIS Fraud Detection dataset. By leveraging AutoGluon’s AutoML capabilities, we automate the entire process from data preprocessing to model selection and optimization, ensuring a high-performing classifier with minimal coding.

Dataset

The dataset comprises both transaction and identity information, making it ideal for building a comprehensive fraud detection model. The key files used in this project are:

• train_transaction.csv — Transaction data with the target label (isFraud).
• train_identity.csv — Identity attributes corresponding to transactions.
• test_transaction.csv — Transaction data for prediction (no labels).
• test_identity.csv — Identity attributes for test transactions.
• sample_submission.csv — A template for submitting the final predictions.

Project Workflow

1. Data Preparation

• Merge train_transaction.csv with train_identity.csv to create the training dataset.
• Similarly, merge test_transaction.csv with test_identity.csv for the test dataset.

2. Model Building

• Use AutoGluon’s TabularPredictor to create a classification model.
• Specify isFraud as the label and use roc_auc as the evaluation metric, given the imbalanced nature of the dataset.

3. Model Training

• Train the model using a subset of the data with AutoGluon’s presets='good_quality' to balance model quality and training speed.

4. Model Evaluation & Inference

• Generate probability scores for the test data and prepare the submission file (my_submission.csv).

5. Submission

• Save and download the submission file for evaluation on Kaggle.

Tutorial link: https://youtu.be/ilewdbDnjTU