The dramatic increase in community-wide infectious disease data - generated by electronic health records, mobile health applications and public health data reporting systems has created opportunities in machine learning (ML) like never before, to assist with predicting outbreaks, monitoring diseases and taking community health-related actions. But due to the sensitive health data, privacy, security, and legal issues are high. The classical centralized method of ML creates a risk of revealing personally identifiable information and can be not in accordance with the new regulations such as GDPR and HIPAA. To overcome these issues, privacy-saving methods, including secure aggregation and differential privacy, have become the key to the implementation of legalization of ML on distributed health data. The review critically reviews the principles, applications and limitations of these techniques as they are applicable to infectious disease analytics. It summarises prior studies on secure aggregation protocols, differential privacy schemes and federated learning designs, demonstrating the contribution each contributes to the privacy of sensitive health information and the utility of models. Also, the review emphasizes the fact that there are certain key challenges, i.e., scalability, the problems of accuracy-privacy trade-offs, and integration with legal frameworks, and indicates the directions that should be followed in the future research to enhance the technical and regulatory compliance. The content of the review is expected to inform the researcher, policymakers, and practitioners on how to create effective, secure, and ethically responsible strategies to community-wide surveillance and analysis of infectious diseases through the use of ML, as it offers a general idea about the current approaches to privacy protection in the context of big data and analytics practice.