Credit Card Fraud (CCF) detection is a major challenge in financial security, especially in detecting unauthorized transactions. As fraudsters' patterns increase, traditional methods face challenges in detecting them. Hence, this paper proposes a Sea Lion - Self-Supervised Network (SL-SSNet) to improve detection accuracy. This research study aims to enhance performance by optimizing data quality feature extraction and achieving better results in fraud detection. The innovative approach for CCF detection using a hybrid optimization model combines the strengths of Sea Lion optimization and Self-Supervised Networks to improve both model accuracy and performance. Initially, the CCF dataset is collected from Kaggle. Then the data goes through a pre-processing phase where irrelevant data points, noise, and low-quality data are removed. The relevant data is selected in the next phase. Feature extraction is performed to select the most important and influential features related to fraud detection. The final phase is prediction and performance evaluation. The results show that the SL-SSNet model performs better than other methods in fraud detection. Specifically, the model achieved 99.98% accuracy, 82.46% precision, 97.23% recall, and 89.97% F1-score. These results prove the effectiveness and robustness of SL-SSNet in detecting fraudulent transactions.

In the rapidly evolving landscape of blockchain technology, accurately classifying blockchain addresses is crucial for enhancing security, tracking illicit activities, and understanding transaction behaviors. This paper presents a novel classification method leveraging a Graph Neural Network (GNN) framework, which utilizes an account balance model to analyze relationships and transactions between blockchain addresses. By constructing a graph representation of the blockchain network, we capture both the structural and functional attributes of addresses. Our approach demonstrates significant improvements in classification accuracy compared to traditional methods, offering a robust solution for identifying various address types, including exchanges, personal wallets, and service providers. Experimental results show that our model effectively distinguishes between different classes of addresses, thereby contributing to the broader field of blockchain analytics and security.

The quick advancement of big data sharing increases the importance of protecting data privacy. Differential privacy appeared in 2006 to become the fundamental privacy-preserving data mining and publication method. Various technologies, including federated learning, homomorphic encryption, and secure multi-party computation, have introduced new solutions for safeguarding data privacy during recent technological developments. Differential privacy is experiencing a comparison test in current technology markets because Google and Apple have integrated it into new products. This chapter begin by examining the differential privacy systems used in traditional centralized data models and their usage in data mining and publishing processes. The chapter investigates modern developments in local differential privacy by examining randomized response Bloom filters and statistical inference tools related to crowdsourcing systems.