Optimising Credit Card Fraud Detection through Machine Learning and Deep Learning with Spatial-Temporal Imbalance Handling - Thesis Abstract

Aug 4, 2026

The sharp rise in online financial transactions during the recent epidemic has increased dependence on digital payments while also expanding financial scams, making robust credit-card fraud detection urgent. The study targets the class imbalance challenge that weakens conventional detection methods and extends the framework with geolocation and temporal signals to reveal trends and irregularities. A combined ML and DL pipeline is developed using multiple data-balancing strategies and spatial-temporal attention, then evaluated with Recall, Precision, F1, ROC-AUC, and Accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages175

Reading Time7 h 21 min

AccessFree to download

Abstract

Problem context: class imbalance and fraud risk

Proposed approach: ML + DL with spatial-temporal handling

Data balancing methods and models evaluated

Evaluation metrics and key findings

Optimising Credit Card Fraud Detection through Machine Learning and Deep Learning with SpatialTemporal Imbalance Handling by Nur Indah Lestari Thesis submitted in fulfilment of the requirements for the degree of Master of Science (Research) in the School of Computing Sciences under the supervision of Associate Professor Walayat Hussain Co-supervisor of Professor Jose Maria Merigo Lindahl University of Technology Sydney Faculty of Engineering and Information Technology December 2024 CERTIFICATE OF ORIGINAL AUTHORSHIP I, Nur Indah Lestari declares that this thesis is submitted in fulfilment of the requirements for the award of Master of Science (Research) degree, in the School of Computing Sciences at the University of Technology Sydney. This thesis is wholly my own work unless otherwise referenced or acknowledged. In addition, I certify that all information sources and literature utilised herein are indicated in the thesis. This document has not been submitted for qualifications at any other academic institution. This research is supported by the Australian Government Research Training Program. Production Note: Signature removed prior to publication. Nur Indah Lestari Date: 12 December 2024 ABSTRACT The sharp rise in online financial transactions during the recent epidemic has led toa record-breaking dependence on digital payment systems. Although this transformation provided a convenient solution, it also led to numerous challenges, such as increased financial scams. Therefore, a robust fraud detection system where individuals can perform online transactions seamlessly is an urgent requirement. This study aims to address the core problems associated with credit-card fraud detection, namely the "class imbalance challenge" that limits the effectiveness of conventional detection methods. The study notes that geolocation and temporal analysis adds an additional dimension to the fraud detection framework, thereby facilitating the identification of trends and irregularities that may otherwise remain undetected. This research aims to develop and validate a novel approach for identifying credit card fraud. The study applied a mix of advanced Machine Learning (ML) and Deep Learning (DL) algorithms, data balancing approaches, and spatial-temporal attention systems to address the research problem. To achieve the objectives, the research first highlights the importance of data-balancing approaches in enhancing model performance within the context of unbalanced datasets. Second, we applied balancing approaches such as Random Over Sampling (ROS), Synthetic Minority Over-sampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), and Random Under Sampling as per the literature analysis, which significantly enhanced the models' capacity to detect fraudulent transactions. Subsequently, we implemented eight different machine learning algorithms, namely Bagging Classifier, Random Forest Classifier, CatBoost, Logistic Regression (LR), Extreme Gradient Boosting (XGBoost) , AdaBoost, Gaussian Naive Bayes (GNB) , and Extra Trees Classifier, along with two deep learning algorithms, namely Gated Recurrent Unit (GRU) and Neural Network (NN) executed. Finally, we conducted the performance measurements, namely Recall, Precision, F1 Score, ROC-AUC Score, and Accuracy, to evidence the success of these strategies. The thesis provides a detailed examination of how machine learning and deep learning models perform in the detection of credit-card fraud, particularly emphasising the effects of data-balancing strategies. The Bagging Classifier and Random Forest Classifier models demonstrate remarkable proficiency, thus attaining outstanding results in all metrics, which indicates their ability to accurately detect fraudulent transactions while maintaining the percentage of false positives at a minimum value. Ensemble approaches outperform simpler models such as LR. Despite exhibiting excellent accuracy, LR fails to

FAQs

What main challenge limits conventional credit-card fraud detection in this study?

The class imbalance challenge, which reduces the effectiveness of conventional detection methods for fraudulent transactions.

How does the study improve the fraud-detection framework beyond standard approaches?

It incorporates geolocation and temporal analysis, adding an extra dimension to identify trends and irregularities that may otherwise remain hidden.

Which evaluation metrics are used to verify the effectiveness of the proposed strategies?

Recall, Precision, F1 Score, ROC-AUC Score, and Accuracy are used to assess performance across models and balancing methods.

Aug 4, 2026

Upload by Clementine