With advances in Cyber-Physical Systems (CPS), privacy-preserving and security issues have attracted substantial attention. A crucial function provided by CPS is anomaly detection on large-scale, complicated, and dynamic data. Physical and network information about the systems for safeguarding original data and identifying cyberattacks is needed in order to develop a reliable privacy-preserving anomaly detection approach. Conventional anomaly detection techniques cannot be directly used to solve these problems because they must deal with the expanding amount of data and need domain-specific expertise. By filtering and choosing key aspects from the original data for improved safety, this research presents a privacy preservation approach for secure anomaly detection. For selecting features, the Mutation Mayfly Algorithm (MMA) has been developed. The proposed program combines key benefits of swarm intelligence and evolutionary algorithms. The usage of MMA in feature selection results from its better accuracy and straightforward structure. Then, a strategy for identifying anomalies based on a Kalman Filter (KF) model and a Gaussian Mixture Model (GMM) has been created to find cyberattacks in CPS. Furthermore, the efficacy of privacy-preserving anomaly detection is being improved through the utilization of a Gaussian Mixture Model (GMM) to convert the noteworthy features into representative characteristics. The present study provides a description of the KF approach, which involves the analysis of the dynamics pertaining to both normal and attack events. The system employs a dynamic thresholding technique to detect anomalous behavior by calculating the lower and upper boundaries of normal activity. The architecture is assessed using two open datasets, UNSW-NB15 for network data and Power System for data on cyber power.