Human mental workload (MWL) is a multifaceted paradigm that has garnered significant attention in various disciplines, including Human Factors and Ergonomics, Neuroscience, and Neuroergonomics. A single, broadly applicable approach for mental workload research hasn’t yet been fully defined, despite an increasing number of studies. The fact that different operational definitions are based on differing theoretical presumptions is one of the causes of this difference. As a result, multiple definitions of mental workload have emerged, which can complement one another. To support future research, it is suggested that a new, inclusive definition of mental workload is adopted. This work provides a summary of recent advancements in machine learning algorithms, including SVM, Random Forest, ANN, SVR, LDA, RNN, Recurrent 3D CNN, and BLSTM-LSTM networks for analyzing mental workload using EEG for both single and cross-task performance. The work begins by introducing the various metrics involved in classifying human mental workload, followed by an outline of the structural principles, characteristics, and different neurophysiological measurements used to estimate MWL. Also, presents an analysis of the challenges involved in measuring different workload levels with better accuracy and discusses possible solutions. Finally, the performance of different machine learning algorithms in classifying workload levels are compared. To establish a framework for mental workload research that is more trustworthy and broadly applicable, it is necessary to adopt a more comprehensive and integrated approach that employs multiple measures and definitions of mental workload.