Fraud Detection System for FASTag Transactions

Project Overview

This project leverages machine learning classification techniques to develop an effective fraud detection system for FASTag transactions. The dataset includes critical features such as transaction details, vehicle information, geographical location, and transaction amounts. The primary objective is to build a robust model capable of accurately identifying instances of fraudulent activity, thereby safeguarding the integrity and security of FASTag transactions.

Project Objectives

Exploratory Data Analysis | Notebook Link

• Peak fraudulent activity times: Most frauds occur at 4 PM, 10 PM, and 6 AM.
• Months with highest fraud incidents: January recorded the highest number of frauds, followed by March.
• Lane with the highest fraudulent activity: Lane B102 experiences the most fraudulent transactions.
• Vehicle types involved in fraud: Large vehicles, particularly SUVs and Vans, are most frequently involved in fraudulent activities. Sedans and Trucks also show significant involvement.

Data Preprocessing | Notebook Link

• Dropped Columns: Removed Transaction_ID, FastagID, Vehicle_Plate_Number, Transaction_Amount, Amount_paid, and Timestamp as they were not relevant for analysis.
• Feature Engineering: Created a State column from Vehicle_Plate_Number to map states accurately. Extracted Month and Time of Day from Timestamp.
• Encoding: Used One-Hot Encoding (OHE) for categorical columns to handle categorical data effectively.
• Scaling: Applied StandardScaler to numeric columns for uniform scaling and preparation for machine learning models.

Model Selection and Comparison | Notebook Link

• Models Evaluated: Logistic Regression, Random Forest Classifier, KNN Classifier, Gradient Boosting Classifier, XGBoost, CatBoost, SVM Classification.
• Performance Evaluation: Compared using F1-score due to dataset imbalance, which balances precision and recall effectively.
• Best Performing Model: KNN Classifier achieved the highest F1-score, indicating its superior ability to balance precision and recall, making it the recommended choice for managing fraudulent activity detection effectively.

Model Training and Evaluation (Hyperparameter Tuning) | Notebook Link

• Hyperparameter Tuning: Employed RandomizedSearchCV to optimize KNeighborsClassifier parameters (n_neighbors, weights, metric) using 3-fold cross-validation.
• Result:
  • Achieved a high recall (98%) for fraud detection, indicating effective identification of actual fraud cases.
  • Precision for fraud (79%) suggests reliable predictions when fraud is predicted.
  • Overall performance metrics include 78% accuracy and a balanced F1-score of 0.70, showcasing effective classification across both fraud and non-fraud cases.

This approach ensures the pipeline is optimized for performance, particularly in detecting fraud, and validates its effectiveness with robust evaluation metrics.

Pipeline Building | Notebook Link

• Built a pipeline using ColumnTransformer for preprocessing (one-hot encoding categorical features, scaling numeric features) and integrated a KNeighborsClassifier for classification.
• Exported the said pipeline and made predictions loading the same.

Results

The KNN Classifier demonstrated the best performance with the following metrics:

• Recall for fraud detection: 98%
• Precision for fraud detection: 79%
• Accuracy: 78%
• F1-score: 0.70

These results highlight the model's ability to effectively identify fraudulent transactions while maintaining a balance between precision and recall.