Fraud Detection Model

DMP: Fraud Detection in Online Transactions DescriptionFraud Detection using Machine Learning Models Context and Methodology Research Domain/Project:This dataset supports a machine learning project focused on detecting fraudulent activities in online financial transactions. Fraud detection is critical in e-commerce, banking, and digital payments to reduce financial losses and enhance trust. The experiment uses structured transaction data, including customer metadata and transaction patterns, to train models that identify fraudulent behavior. Purpose:The primary objective is to develop and compare machine learning models for their ability to detect fraudulent transactions accurately. These insights can inform the deployment of real-time fraud prevention systems in financial platforms. Creation Process:The dataset is synthetically generated to simulate real-world transactional behavior. It includes transaction amount, timestamp, customer location, payment method, and historical transaction count. A small percentage of transactions are labeled as fraudulent to represent realistic class imbalance. The dataset design adheres to domain standards and was informed by patterns observed in financial fraud datasets publicly available on Kaggle and academic research. Technical Details Dataset Structure:The dataset comprises 5,000 records, each representing a financial transaction. The following features are included: Transaction Amount Timestamp Customer Location Payment Method Historical Transactions Fraud Label (target variable) These features are selected to enable the detection of anomalies and patterns associated with fraud. Software Requirements: Python: Used for all data processing and modeling scikit-learn: For training and evaluating machine learning models Pandas and NumPy: For data handling and preprocessing Matplotlib/Seaborn: For visualizing results DBRepo Authorization: Required to load datasets via the DBRepo API for reproducibility and citation Additional Resources Evaluation Metrics: Confusion matrices, classification reports, and ROC curves are generated to compare model performance Trained Models (.pkl files): Saved versions of trained models for inference without retraining. Metadata is provided in FAIR4ML format Documentation and Code: A complete Jupyter notebook includes data loading, preprocessing, model training, evaluation, and saving outputs. Code is described using CodeMeta metadata