Fraud Detection

Objective

The objective of this project is to develop a model that can accurately detect fraudulent transactions based on features such as transaction amount, customer ID, terminal ID, and more.

Dataset

The dataset includes various features for transactions that help in identifying fraud cases. Key columns include transaction identifiers, amounts, timestamps, and indicators of whether a transaction was fraudulent.

Features:

• TRANSACTION_ID: Unique identifier for each transaction
• TX_DATETIME: Date and time of the transaction
• CUSTOMER_ID: Unique identifier for each customer
• TERMINAL_ID: Unique identifier for each terminal
• TX_AMOUNT: Amount of the transaction
• TX_TIME_SECONDS: Time in seconds of the transaction during the day
• TX_TIME_DAYS: Time in days since the start of the dataset
• TX_FRAUD: Target variable indicating whether the transaction was fraudulent (1) or not (0)
• TX_FRAUD_SCENARIO: Code representing different fraud scenarios

Additional Features:

Derived features such as:

• TX_DAY, TX_MONTH, TX_YEAR, TX_HOUR: Extracted from TX_DATETIME for more granular time-based analysis

Approach

Data Preprocessing:

• Handling Missing Values: Verified no missing values were present.
• Feature Engineering: Converted categorical columns (CUSTOMER_ID, TERMINAL_ID) to numeric, and extracted day, month, year, and hour from TX_DATETIME.
• Scaling: Standardized numerical features using StandardScaler to improve model performance.

Splitting Data:

The dataset was split into training and testing sets with a 70-30 split.

Modeling

A RandomForest Classifier was employed for detecting fraud. The model was trained and evaluated on scaled features.

Model Evaluation:

• Accuracy: Achieved an accuracy of 100% on the test set.

• Confusion Matrix: precision recall f1-score support

   0       1.00      1.00      1.00    521843
   1       1.00      1.00      1.00      4404
  

accuracy 1.00 526247

Model Saving

The trained RandomForest model was saved as fraud_detection_model.joblib using joblib for easy deployment.

Running the Fraud Detection System

1. Clone the repository and ensure all dependencies are installed.
2. Load the model file fraud_detection_model.joblib.
3. Use the model to predict fraud on new transactions by scaling the input features as done during training.

Visualization

A heatmap of the confusion matrix was generated to illustrate the model's performance.

Deployment

The saved model can be integrated into a real-time system to identify and flag potential fraudulent transactions based on incoming data.