| dc.description.abstract | In today's digital age, fraudulent activities, especially on online money transfer platforms, have grown significantly, requiring the development of more sophisticated detection methods. This study proposes a new fraud detection method integrating Louvain-Coloring, PageRank, Degree Centrality and Random Forest algorithms. This approach uses the concept of network analysis to identify suspicious patterns and relationships in large-scale datasets. First, the Louvain algorithm is used to divide the network into different communities, which are considered potential “clusters” of fraudulent activity. PageRank is then used to identify important nodes in these clusters based on their connection strength and frequency. Centrality enhances this process by identifying highly connected nodes, which often indicate coordinated fraud. These measurements provide a comprehensive understanding of data structure and potential fraud hotspots, which are then used as features for the Random Forest algorithm. Random Forest is a powerful machine learning algorithm known for its reliability and accuracy, making it ideal for classifying nodes as "fraudulent" or "non-fraudulent" based on characteristics it draws from. The proposed method was validated using online money transfer transaction data, which contained 33,491 correct transactions and 241 fraudulent transactions. The proposed method provides a maximum accuracy value of 0.89 and a prediction of 0.93 and an AUC of 0.96 and recommends 1 transaction with a prediction value greater than 0.91 so that transactions classified as fraudulent became 242 transactions. | en_US |
