FRAUD DETECTION IN BANKING FINANCIAL TRANSACTIONS USING SVM WITH THE APPLICATION OF A COMBINATION OF SMOTE AND TOMEK LINKS SAMPLING METHODS

• Main Author: Ayu Novita Prahasta Dewi, Kadek
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/80968
• Institution: Institut Teknologi Bandung

The field of information technology is expanding at a rapid rate, and with the aid of electronic devices, community activities2particularly in the financial sector2 are thought to help, but they also carry some risk of encouraging fraudulent financial transactions, which could indirectly increase the number of fraud cases. Fraudsters continue to find loopholes in financial transactions and by using technology, they can commit these crimes. One of the ways to detect fraud more quickly and accurately is fraud detection by applying machine learning which is revealed from a number of previous studies. The main problem in fraud detection is highly imbalanced data where genuine transactions dominate over fraudulent transactions. Machine learning models learn from data and create patterns, so if the data is unbalanced then the model cannot identify fraud correctly. The Support Vector Machine (SVM) technique is used in this study to tackle the challenge and continue the classification process using a combined sampling method called SMOTE and Tomek Links. The findings demonstrated that the SVM algorithm's kinerjance can be enhanced by using the combined sampling method of SMOTE Tomek Links. This can be shown in the high accuracy value before carrying out the resampling technique, namely 94% for the X Bank Dataset and 95% for the Banksim Dataset, however this model was not successful in detecting fraud due to the class imbalance in these two datasets. After resampling and hyperparameter tuning, it shows that the resulting Accuracy, Recall, Precision and F1-Score values are 0.91, 0.48, 0.36, 0.41 for the X Bank Dataset and 0.96, 0.96, 0.53, 0.68 for the Banksim dataset.