This repository contains the code and documentation for the Fraud Detection Model project. The goal of the project is to develop a machine learning-based system capable of identifying fraudulent financial transactions using Logistic Regression and Random Forest algorithms. The project showcases key aspects of imbalanced classification, data preprocessing, and model evaluation.
The Fraud Detection Model project applies machine learning techniques to detect fraudulent transactions from a financial dataset. The model is trained to classify transactions into fraud and non-fraud categories. Given the highly imbalanced nature of the dataset (with a fraud ratio of just 1.44%), special attention was given to handling the class imbalance and evaluating the model's performance on the fraud class.
- Number of Rows: 20,000
- Number of Features: 30 (e.g., transaction amount, time, etc.)
- Fraud Ratio: 1.44% (indicating highly imbalanced data)
The dataset is synthetic, designed specifically for fraud detection challenges, and includes both legitimate and fraudulent transactions.
- ROC AUC: 0.5504
- Confusion Matrix (Test Data):
[[4928 0]
[ 72 0]]
- Fraud Class (1) Performance:
- Precision: 0.0000
- Recall: 0.0000
- F1-Score: 0.0000
Logistic Regression serves as a baseline model. Despite achieving a moderate ROC AUC, it struggled to detect fraudulent transactions effectively due to class imbalance.
- ROC AUC: 0.4509
- Confusion Matrix (Test Data):
[[4928 0]
[ 72 0]]
- Fraud Class (1) Performance:
- Precision: 0.0000
- Recall: 0.0000
- F1-Score: 0.0000
While Random Forest showed a slight improvement in ROC AUC, the model similarly struggled with detecting fraud, highlighting the need for further improvements in data handling and model optimization.
-
Class Imbalance: Both models showed limitations in detecting fraud due to the highly imbalanced nature of the dataset, with fraud accounting for only 1.44% of the total transactions.
-
Model Performance: Despite reasonable ROC AUC scores, the models performed poorly in detecting fraud (class 1), indicating that additional techniques such as SMOTE (Synthetic Minority Over-sampling Technique) or ensemble methods are needed to improve performance on imbalanced datasets.
-
Evaluation Metrics: While ROC AUC was used to evaluate the overall performance, other metrics such as Precision, Recall, and F1-Score should be given more focus for the fraud class to ensure real-world applicability.
The dataset had a highly imbalanced class distribution. Here’s a visualization of the class balance:
The confusion matrix for the Random Forest model, showing the model’s limited ability to detect fraudulent transactions:
To get started, clone the repository to your local machine:
git clone https://github.com/your-username/Fraud-Detection-Model.gitMake sure to install the required libraries by running:
pip install -r requirements.txt- Open the Jupyter notebook
fraud_detection.ipynbto view the implementation. - Run the notebook cells to train the model, evaluate its performance, and generate visualizations.
- The models will be trained on the synthetic dataset, and you will see metrics like ROC AUC, Confusion Matrix, and other evaluation results for both models.
For a detailed explanation of the methodology, evaluation, and conclusions, refer to the project report (fraud_project_report.md). The report provides in-depth insights into the model development process.
- Resampling Techniques: Implement techniques like SMOTE (Synthetic Minority Over-sampling Technique) to better handle the class imbalance in the dataset.
- Model Improvement: Experiment with more advanced models such as XGBoost or Neural Networks to improve fraud detection accuracy.
- Hyperparameter Tuning: Fine-tune the model parameters using grid search or random search to optimize performance.
- Feature Engineering: Further feature selection and engineering can improve model performance by removing irrelevant or highly correlated features.
- Thanks to Zidio for providing the opportunity to work on this project and gain hands-on experience in machine learning and fraud detection.
- Project Lead: Tejas Solanki
- Contributors: Tejas Solanki, Sonu Kumar, Nirgun Joshi, Rupak Lipane, Rony Varghese
- Special Thanks: Open source libraries (Yahoo Finance, pandas, scikit-learn, TensorFlow, Streamlit, etc.)
This project is licensed under the MIT License - see the LICENSE file for details.
This GitHub repository showcases a Fraud Detection Model designed to identify fraudulent financial transactions. The project provides valuable insights into working with imbalanced datasets and applying machine learning algorithms in real-world applications. Future improvements will focus on enhancing model performance by addressing class imbalance and experimenting with more advanced techniques.