Deep Learning-Based Anomaly Detection in Cyber-Physical Systems

Abstract

Although anomaly identification is essential for identifying infrequent occurrences, these methods frequently fall short in differentiating anomaly from noise. While they are related, anomaly detection is different from noise removal. While anomaly detection is a component of data analysis, noises are undesired objects that impede the process. Although noise can be represented as a weak type of anomaly, this does not imply that all noise is abnormal. The elimination and identification of noise is crucial for anomaly detection, even though it is true that the distinction between the two is not exact. To tackle the challenge of anomaly detection, researchers have come up with various machine learning and data-mining algorithms. However, these methods come with their own set of issues, like forcing data points into specific clusters, class dominance, local convergence, and being overly sensitive to how clusters are chosen. Additionally, it's been observed that the same technique can produce different results when used on different datasets. Identifying abnormalities in cloud servers is vital for maintaining data confidentiality, especially with the rise of massive data storage and access in cloud computing. The security systems developed in the early stages of this study are more effective than previous access control methods. However, those earlier methods assumed that cloud servers could be trusted. In reality, the server in this system might be untrustworthy, as it often manipulates encrypted data and can be caught doing so.