Voice stress analysis (VSA) has received a lot of attention as a non-invasive tool for detecting deceit and emotional states by analysing speech patterns. Machine Learning (ML) has emerged as a potent technique for improving the precision and efficiency of VSA approaches. This paper provides an in-depth examination of the applications and improvements in voice stress analysis using ML approaches. This study explores various ML algorithms and feature extraction approaches typically used in VSA, highlighting their strengths and limitations. Furthermore, this approach discusses the problems and prospects in the field of VSA, emphasising the potential for incorporating cutting-edge ML methods to improve accuracy, dependability, and real-time applications.