An end-to-end machine learning project to detect fraudulent credit card transactions using an imbalanced dataset. This project demonstrates how predictive modeling in R (including logistic regression, SMOTE, and XGBoost) and interactive dashboards in Power BI can work together to support financial risk management in real-world settings.
Credit card fraud is a growing threat in the digital economy. The challenge lies in identifying rare fraudulent transactions hidden among thousands of legitimate ones. This project uses predictive modeling and business intelligence to:
- Classify transactions as fraudulent or legitimate
- Deal with class imbalance (only ~0.17% are fraud)
- Uncover patterns and high-risk signals in transactional behavior
- Compare multiple modeling approaches for optimal performance
- Communicate findings through an executive-friendly Power BI dashboard
- Source: Kaggle - Credit Card Fraud Detection
- Transactions made by European cardholders over a two-day period in 2013
- 284,807 transactions, with only 492 labeled as fraud
- Features:
V1–V28: PCA-transformed numerical featuresAmount: Transaction value (in Euros)Time: Seconds since the first transactionClass: Target variable (1= Fraud,0= Legit)
| Category | Tools Used |
|---|---|
| Language | R (tidyverse, caret, corrplot, etc.) |
| Modeling | Logistic Regression, Random Forests, SMOTE |
| Evaluation | pROC, PRROC, yardstick, confusionMatrix |
| BI Dashboard | Power BI |
| Visualization | ggplot2, corrplot, Power BI |
| Reporting | RMarkdown, CSV summaries |
| Documentation | MS Word |
- Visualized amount distributions and transaction times
- Compared fraudulent vs. legitimate transaction characteristics
- Standardized numerical features (
Amount,Time) - Created additional features (e.g., time-based indicators)
- Handled extreme class imbalance using SMOTE oversampling
- Baseline Logistic Regression: Established performance baseline
- SMOTE-enhanced Logistic Regression: Addressed class imbalance
- XGBoost: Advanced gradient boosting for improved performance
- Comprehensive model evaluation using precision, recall, F1-score, ROC, and Precision-Recall curves
- Created interactive KPIs and charts summarizing fraud distribution
- Visualized transaction value trends and model predictions
- Designed an intuitive interface for fraud analysts and stakeholders
⚡ Note: The dashboard is a work in progress — additional fraud trend visualizations and analyst-focused KPIs are being developed.
📦 credit-card-fraud-detection/
├── data/
│ ├── raw/
│ │ └── creditcard.csv
│ └── processed/
│ ├── creditcard_clean.csv
│ ├── creditcard_model_ready.csv
│ ├── fraud_summary_for_powerbi.csv
│ ├── logistic_predictions.csv
│ ├── logistic_smote_predictions.csv
│ ├── logistic_metrics.csv
│ ├── logistic_smote_metrics.csv
│ ├── smote_data.RData
│ ├── train_data.csv
│ └── test_data.csv
├── outputs/
│ ├── models/
│ │ ├── logistic_model.rds
│ │ ├── logistic_model_smote.rds
│ │ └── xgboost_model.rds
│ ├── plots/
│ │ ├── amount_distribution.png
│ │ ├── log_amount_distribution.png
│ │ ├── time_distribution_by_class.png
│ │ ├── amount_boxplot_by_class.png
│ │ ├── density_V1.png ...
│ │ ├── correlation_matrix.png
│ │ ├── feature_correlation.png
│ │ ├── roc_curve.png
│ │ ├── pr_curve.png
│ │ ├── xgboost_roc_curve.png
│ │ ├── xgboost_pr_curve.png
│ │ ├── xgboost_feature_importance.png
│ │ └── model_comparison.png
│ └── reports/
│ ├── logistic_confusion_matrix.txt
│ ├── logistic_smote_confusion_matrix.txt
│ ├── xgboost_report.txt
│ ├── model_comparison.csv
│ └── model_comparison_detailed.csv
├── scripts/
│ ├── 00_master_script.R
│ ├── 01_data_load.R
│ ├── 02a_eda.R
│ ├── 02b_eda_features.R
│ ├── 03b_preprocessing.R
│ ├── 03a_feature_selection_scaling.R
│ ├── 04_model_data_walkthrough.R
│ ├── 05a_modeling_data_split.R
│ ├── 05b_model_baseline.R
│ ├── 05c_model_improvement_smote.R
│ ├── 05d_model_improvement_xgboost.R
│ └── 06_model_comparison.R
└── logs/
└── master_script_log.txt
To run the entire project workflow from data loading to model training and output generation, execute the master orchestrator script:
# From the project root directory
source("scripts/00_master_script.R")
- The master script automatically installs required packages and handles all dependencies
- All file paths are managed using the
herepackage for reproducibility across systems- Execution progress and timing information is logged to
outputs/logs/master_script_log.txt- The Kaggle dataset (
data.csv) must be placed indata/raw/before execution
| Model | AUC | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|
| Logistic Regression (Baseline) | - | 0.873 | 0.579 | 0.696 | 0.999 |
| Logistic Regression (SMOTE) | 0.977 | 0.794 | 0.794 | 0.794 | 0.999 |
| XGBoost | 0.982 | 0.944 | 0.794 | 0.863 | 0.999 |
- XGBoost achieved the best overall performance with highest AUC (0.982) and F1-score (0.863)
- SMOTE significantly improved recall (from 0.579 to 0.794) while maintaining high precision
- Fraudulent transactions often have smaller or atypical amounts
- Certain PCA features show strong separation by class
- Precision-recall trade-off is critical in low-fraud environments
- The Power BI dashboard enables real-time fraud monitoring and investigation
💡 Power BI dashboard development is ongoing — the current version includes fraud distribution summaries, with future updates planned for advanced drilldowns and real-time monitoring views.
- Finalize and expand the Power BI dashboard (advanced drilldowns, fraud KPIs)
- Deploy the best model (XGBoost) as a real-time scoring API
- Implement automated model retraining pipeline
- Integrate additional data sources (geolocation, merchant metadata)
- Develop an alert system for high-risk transactions
- Explore deep learning approaches (autoencoders, LSTM networks)
- Implement model monitoring for concept drift detection
