IMPLEMENTATION OF NAIVE BAYES-BASED MACHINE LEARNING ALGORITHM IN THE FRAUD DETECTION SYSTEM

• Main Author: Nurulhaqi Syahidah, Aisyah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/49929
• Institution: Institut Teknologi Bandung

Technological developments have an impact on financial transaction methods in the current era. The appearance of mobile transaction services such as e-banking, m-banking, digital payment, and others has become usual in the community. One service that is currently developing is mobile money. This service makes it easy for users to access various financial transactions from cellular phones, for example, are Gopay, OVO, and Dana. Behind the convenience obtained by the user, some risks must be managed so as not to cause harm to both parties, namely users and financial service providers. One of the risks is a fraud that can be observed from the transactions that occur. In this final project, a machine learning algorithm based on Naive Bayes is used, which consists of two algorithms, namely Gaussian Naive Bayes and Categorical Naïve Bayes. The algorithm is implemented through five stages, which are understanding business needs, understanding data, pre-processing data, optimizing parameters and modeling, and evaluating. In the pre-processing data stage, data transformation is performed by the algorithm to be created, the handling of the data imbalance with a combination of oversampling and undersampling methods using the SMOTE algorithm and the RandomUndersampler. At the parameter optimization and modeling stage, the random search method is used with the k-fold crossvalidation algorithm. The evaluation is carried out using four different metrics, which are recall, specificity, precision, and F1 score. The results of this study indicate that when the algorithm is applied to the PaySim test data, the Gaussian Naive Bayes algorithm provides better performance than the Categorical Naive Bayes algorithm. The Gaussian Naive Bayes algorithm can achieve recall 0.926540, fpr 0.190328, specificity 0.809672, auc score 0.868106, precision 0.014658, and F1 score 0.028859.