Maximal oxygen uptake (VO2 max) is the maximum amount of oxygen attainable by a person during exercise. VO2 max is used in different domains including sports and medical sciences and is usually measured during an incremental treadmill or cycle ergometer test. The drawback of directly measuring VO2 max using the maximal test is that it is expensive and requires a fixed and controlled protocol. During the last decade, various machine learning models have been developed for VO2 max prediction and numerous studies have attempted to predict VO2 max using data from submaximal and non-exercise tests. This article gives an overview of the machine learning models developed over the past five years (2016–2021) for the prediction of VO2 max. Multiple linear regression, support vector machine, artificial neural network and multilayer perceptron are some of the techniques that have been used to build predictive models using different combinations of predictor variables. Model performance is generally assessed using correlation coefficient (R-value), standard error of estimate (SEE) and root mean squared error (RMSE), computed between ground truth and predicted values. The findings of this review indicate that models using ANN typically outperform other machine learning techniques. Moreover, the predictor variables used to build the model have a large influence on the model's predictive performance.