The data analyst aspires to train their ML model, such as k-means with the IoMT data. However, the data owners are not confident enough to share their private data with the data analyst for security and privacy concerns. This study proposes privacy-preserving k-means based on Paillier. All transactions are recorded in a distributed, immutable ledger for authenticity and secure data sharing among the participants. Three medical datasets are used and performance analysis exhibits that secure k-means achieves accuracies of 94.95%, 81.88%, and 78.10% on BCWD, HDD, and DD dataset, where standard techniques provide 96.60%, 81.00%, and 77.00%, respectively. On the other hand, secure k-means takes 2200 s, 1500 s, and 2605 s, where the standard method takes 3357 s, 2534 s, and 3709 s on BCWD, HDD, and DD datasets, respectively. Therefore, secure k-means can protect the privacy of the data owners, achieves almost comparable accuracy to the conventional methods, and outperforms them in time consumption.