Secure K-Means Clustering Using Partially Homomorphic Encryption For Privacy Preservation

Machine learning (ML) has emerged as a pillar across many industries, improving efficiency and productivity in a wide range of applications. As more organizations utilize cloud services to perform computations, though, there is an increasing risk of revealing sensitive information in the process. Traditional K-Means clustering algorithms operate directly on raw data, which, when transmitted to the cloud or third-party servers, poses the risk of privacy compromise. This study fills this gap by suggesting a secure clustering technique based on Partial Homomorphic Encryption (PHE), utilizing the Cheon-Kim-Kim-Song (CKKS) fully homomorphic encryption in partial mode. This method guarantees data privacy while allowing clustering operations. The suggested algorithm was tested on eight different datasets, and the outcomes were compared to the conventional K-Means algorithm. The experiment proves that the suggested method retains data privacy well without compromising the performance of clustering. The results indicate that the method provides a privacy-preserving, secure option for ML usage, especially for cloud-based environments. The research contributes to constructing secure machine learning methods that ensure privacy and efficiency in computation, representing a meaningful improvement in privacy-preserving data analysis.