Phasor measurement units and wide area measurement systems are becoming more and more popular due to their capability to record operational data with high sampling rates. By storing and processing this large amount of data, faster and more reliable approaches can be developed that overcome some of the drawbacks of traditional methods, such as response speed and accuracy. Many research studies use pattern recognition methods and machine learning techniques to predict the stability of a system following disturbances (unpredicted events). This paper aims to deliver a review of research work carried out in recent years for the assessment of transient stability by focusing particularly on the machine learning techniques. Specifically, supervised and unsupervised learning techniques such as support vector machines, neural networks including hybrid and ensemble models. Moreover, the methodologies that the researchers followed to develop such models including data generation, feature selection and validation are also reviewed.