A machine learning project to detect fraudulent credit card transactions and analyze the cost-benefit of model deployment for a financial services firm.
To build a fraud detection system using machine learning for Finnex, a US-based financial services company, with the goal of reducing fraud losses and improving customer trust. The project includes financial analysis to demonstrate the practical impact of deploying such a model.
- Highly imbalanced dataset: Only 0.57% of transactions are fraudulent.
- Binary classification problem with
is_fraudas the target variable. - Features include transaction details, time, location, customer demographics.
- Random Forest Classifier
- ADASYN (Adaptive Synthetic Sampling) to address class imbalance.
- Manual hyperparameter tuning.
- Evaluation using metrics such as accuracy, precision, recall, and F1-score.
- Fraudulent transactions are more common on weekends and between 10 PM to 3 AM.
- Female customers account for ~55% of total transactions and may be slightly more vulnerable.
- Fraud detection systems can be enhanced using a second layer of authentication.
We evaluated the cost impact of the model by comparing:
| Scenario | Formula |
|---|---|
| Before Model | Average Fraud Amount * Avg. Monthly Fraud Count |
| After Model | 1.5 * True Positives + Average Fraud Amount * False Negatives |
📈 Significant savings observed when the model is deployed.
Details: See Cost_Benefit_Analysis.xlsx
| File | Description |
|---|---|
Abhishek_Fraud_Detection_code.ipynb |
Jupyter Notebook with all preprocessing, model training, evaluation, and analysis |
Credit_Card_Fraud_Detection.pptx |
Executive presentation summarizing the project |
Cost_Benefit_Analysis.xlsx |
Excel sheet evaluating the financial benefit of deploying the fraud detection model |
Watch the 7-minute presentation
- Python
- Pandas, NumPy, Scikit-learn, Imbalanced-learn
- ADASYN
- Jupyter Notebook
- Excel, PowerPoint
Abhishek Kunbhare
- Clone the repo:
git clone https://github.com/abhishek010314/credit-card-fraud-detection.git