This notebook presents a thorough analysis and detection of fraud in online payment transactions. It aims to utilize advanced data processing techniques and machine learning to identify fraudulent activities, leveraging a rich dataset of financial transactions.

The dataset comprises detailed records of online payment transactions over a period of 30 days, simulating typical activities within financial services, particularly focusing on emerging mobile money transactions domains. Key features include:

• step: Maps a unit of time in the real world; 1 step is 1 hour of time. Total steps: 744 (30 days simulation).
• type: Types of transactions (CASH-IN, CASH-OUT, DEBIT, PAYMENT, TRANSFER).
• amount: Amount of the transaction in local currency.
• nameOrig: Customer who started the transaction.
• oldbalanceOrg: Initial balance before the transaction.
• newbalanceOrig: New balance after the transaction.
• nameDest: Customer who is the recipient of the transaction.
• oldbalanceDest: Initial balance of the recipient before the transaction.
• newbalanceDest: New balance of the recipient after the transaction.
• isFraud: Transactions made by fraudulent agents within the simulation.
• isFlaggedFraud: Flags large illegal transfers.

• Renaming columns for clarity.
• Handling outliers and data quality issues using statistical methods.
• Feature engineering to enhance model performance.
• Data standardization and resampling to address imbalanced classes.

• Transactional Analysis: Analyzing total transaction amounts by type, identifying high-frequency accounts, and observing post-transaction balance changes.
• Fraud Analysis: Focused on relationships between transaction amounts and fraudulent activities, including flagged transactions.
• Temporal Patterns: Examining the timing of transactions to detect any patterns related to fraud occurrences.
• Distribution Analysis: Comparing the distribution of amounts in fraudulent versus non-fraudulent transactions.

• Categorical Encoding: Transformation of categorical transaction types using LabelEncoder.
• Dropping irrelevant features: Such as SenderAccountId and RecipientAccountId.

• Random Forest Classifier used with hyperparameter tuning via GridSearchCV.
• Model Performance: Evaluated using accuracy, with details on the best parameters found and test accuracy results.

1. Clone the GitHub repository:

   git clone [repository-url]

2. Navigate to the notebook directory:

   cd [repository-directory]

3. Run the Jupyter Notebook:

   jupyter notebook [notebook-name].ipynb

• Python 3.x
• Jupyter Notebook
• Libraries: pandas, numpy, scikit-learn, matplotlib, seaborn

Contributions to this project are welcome. Please fork the repository, make your changes, and submit a pull request.

This project is released under the MIT License. See the LICENSE file in the repository for more details.

If you have any questions or would like to discuss this project further, please contact the repository owner or submit an issue in the GitHub repository.