Smart Credit Card Fraud Detection using Machine Learning

This thesis addresses the increasing issue of fraud resulting from technological advancements affecting both customers and fraudsters, while law enforcement agencies face challenges of inadequate case prioritization; data overload and slow investigative analysis. The central objective of this thesis revolves around strategies to counter these challenges, emphasizing the urgency of rapid response to major alert cases that can lead to widespread impacts. The authors' experience in financial fraud detection and credit card crime investigation within the police department has inspired this research, guiding the approach toward practical solutions for everyday operational challenges. The significance of this thesis lies in its potential contributions to law enforcement by addressing existing gaps in fraud detection and investigation. The methodology will involve employing a range of emerging big data analytics tools to create the proposed comprehensive solution. Initial phases will focus on data mining and exploratory data analysis (EDA) and a new credit scoring feature to be added to evaluate customers' credibility based on their associated data. The machine learning component will consist of two phases: first, utilizing clustering algorithms for effective case segmentation; and second, implementing classification algorithms on the segmented cases that exhibit higher fraud risk and impact. This dual-phase approach is designed to ensure swift processing of large data sets. Power BI will ultimately be leveraged to construct a dashboard that integrates an alert mechanism for new fraudulent cases, monitors behaviors across different customer segments, tracks the progress of older cases, and aids in the performance evaluation of investigative efforts through key performance indicators (KPIs). The research scope is defined by the dataset utilized, which comprises 32 features related to customer demographics, transaction histories, payment behaviors, and characteristics contributing to fraud detection. Anticipated limitations include restricted access to real-case data from law enforcement agencies due to privacy and security regulations. Conclusions drawn from the research highlight the effectiveness of machine learning and visualization in assisting law enforcement with fraud investigations. By employing clustering algorithms and predictive models, the research addressed key challenges such as data overload and case prioritization. The integration of a visual dashboard further optimized decision-making and resource allocation. These findings offer a practical framework for improving fraud detection efficiency and support the continued development of more advanced systems in the future.