DEVELOPMENT OF MACHINE LEARNING-BASED FRAUD DETECTION SYSTEM USING ARTIFICIAL NEURAL NETWORK ALGORITHM

• Main Author: Mahareksa, Alfiansyah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56155
• Institution: Institut Teknologi Bandung

As time goes on, technology will also develop to be able to support existing needs and challenges. One of them is in the financial sector. For example, just making transactions can be made easier by using mobile money. By using mobile money, transactions that usually take a long time will become easier and faster for users. But unfortunately there are always risks that can occur for each of these conveniences. One of the risks is fraud by irresponsible parties or can be referred to as fraud. Therefore, machine learning will be used to build a fraud detection system by implementing the Artificial Neural Network algorithm which has an artificial nervous system development method to be able to detect which transactions are fraudulent and which are not fraudulent. This development is carried out through five stages in accordance with CRISP-DM. Namely, Business Understanding, Data Understanding, Data Preparation, Modelling, and Evaluation. In the Data Preparation stage, unbalanced money data is handled by undersampling and selecting only the relevant features from the dataset. In the Modelling stage, the Optuna method is used to find optimal hyperparameters by using k-fold cross validation and data backtesting to validate the model created. Then the evaluation is carried out using several parameters, namely, false positive rate (FPR), area under curve, recall, precision, and F1 Score. The results of the ANN model that are made provide FPR performance of 0.078%, area under curve 97.65%, recall 95.34%, precision 99.8%, and F1 score 97.52%.