A Systematic Review of Centralized and Decentralized Machine Learning Models: Security Concerns, Defenses and Future Directions

Abstract

Models are the heart of machine learning as they represent the end product of the learning process and help in making predictions. With the widespread adoption of machine learning and the ever-increasing security concerns especially when dealing with sensitive data like healthcare and financial data, the security of machine learning models has become expedient to guarantee the privacy of training data and ensure the continuous acceptance and adoption of machine learning in fields involving sensitive data. This study presents a systematic review of centralized and decentralized machine learning models; security challenges, common application areas and defence mechanisms implemented to curb the security threats in centralized and decentralized machine learning models. Also, we propose future directions and avenues for research and development to improve the security, performance and resilience of centralized and decentralized machine learning models.