Data Source : The dataset can be found here.
For the fraud detection model, we typically use supervised learning algorithms suited for classification tasks. Here’s an example using a Random Forest Classifier:
Model Choice: Random Forest was chosen due to its robustness and ability to handle imbalanced datasets and its capability of providing feature importance scores.
Data Preparation: The dataset was split into training and validation sets, and SMOTE was used to handle class imbalance.
Feature Engineering: New features were created to capture potential patterns indicative of fraud.
Model Training: The model was trained on the resampled training set.
Evaluation: The model was evaluated using precision, recall, F1-score, and AUC-ROC.
-> Pattern of Fraudulent Transactions: It has been observed that fraudulent transactions often occur in pairs. Specifically, a fraudulent transfer is followed immediately by a corresponding cash-out transaction.
-> Transaction Consistency: For instance, if there is a transfer transaction (T1) of a certain amount, such as 181, it is immediately followed by a cash-out transaction of the same amount.
-> Timing and Amounts: Monitoring the timing and amounts of transactions can provide further insights. Transactions occurring in quick succession, especially for the same amount, should be flagged for further investigation.
-> Structured Fraud Behavior: The consistent pairing of transfer and cash-out transactions indicates a structured and deliberate pattern of fraud. Fraudsters seem to transfer funds and then quickly withdraw them, aiming to minimize the risk of detection and maximize the speed of asset liquidation.
-> Initiation of Fraudulent Transactions: Every fraudulent transaction starts from a customer account, denoted by 'C'. This means that the initial transfer in a fraudulent sequence always originates from a customer account.
-> Recipient Accounts: The recipients of these fraudulent transactions are also customer accounts, denoted by 'C'.
-> Originating Account Type: Knowing that fraudulent transactions always start from customer accounts provides a key focal point for fraud detection systems. Enhanced monitoring and security measures can be implemented specifically for transactions initiated by customer accounts.
-> Uniformity in Recipient Accounts: Since both the initiators and recipients of fraudulent transactions are customer accounts, fraud detection systems can be fine-tuned to monitor customer-to-customer transactions more closely. This uniformity simplifies the detection rules and focuses the analysis on a specific subset of transactions.