Cloud-based healthcare systems offer significant advantages in scalability and flexibility, yet integrating machine learning (ML) into these systems introduces critical challenges, particularly concerning data confidentiality. Unauthorized access to sensitive medical information can lead to privacy breaches and other severe consequences. This paper presents an optimized framework focusing on advanced encryption methods to address this critical issue. In particular, we suggest using homomorphic encryption, which enables processing of data without decryption, preserving confidentiality throughout computation. By reviewing current methodologies and identifying potential vulnerabilities, we demonstrate that homomorphic encryption effectively ensures data security while enabling scalable ML applications. Our evaluation shows that this solution mitigates confidentiality risks and supports efficient data processing, paving the way for more secure and scalable ML implementations in cloud-based healthcare systems.