| Abstract: | This study investigates how deep learning can be applied to techniques to detect and mitigate
competency certification fraud within the Petroleum and Energy Authority (PEA). It addresses
major difficulties in detecting fraud, including counterfeit certifications, varied data formats, and
inconsistencies in records. Despite advancements in fraud detection, existing traditional methods
struggle with scalability, adaptability to evolving fraud patterns, and the ability to effectively
process large-scale tabular data. To bridge these gaps, a tailored deep learning framework has
been designed to meet the PEA's specific requirements, ensuring accurate and efficient fraud
detection.
The proposed system operates in three primary phases: data preprocessing, feature
extraction, and fraud identification. Utilizing a preprocessed dataset, advanced models like
TabNet and DNN are implemented to achieve high accuracy in identifying fraudulent
certifications. TabNet is chosen due to its ability to efficiently process tabular data, its
interpretable decision-making process, and its capacity to capture complex feature dependencies.
Meanwhile, DNN is employed for its deep hierarchical feature learning, allowing it to recognize
intricate fraud patterns within certification data.
Data preprocessing strategies, including normalization, handling of missing values, and
feature scaling, enhance data quality and optimize model performance. By analyzing the
relationships among certification attributes, the system identifies anomalies and uncovers fraudrelated patterns. The framework was trained and validated on a dataset of 9,000 records,
augmented to improve model robustness. The methodology achieved a fraud detection accuracy
of 95.3% (to be updated with actual results), demonstrating its effectiveness in detecting
fraudulent certifications. This system offers a significant advancement in strengthening the
integrity and reliability of the PEA’s certification processes |
