Abstract Mental workload (MWL) is a critical factor influencing human performance, particularly in complex and dynamic environments such as driving, aviation, and healthcare. Accurate MWL estimation is essential for optimizing system design and enhancing user performance. With advancements in technology, automated vehicles are becoming a reality, yet the cognitive processes involved in sharing control between drivers and vehicle systems remain insufficiently understood. This study investigated the mental workload imposed by these cognitive processes, focusing on the demands of primary executive functions. To achieve this, an experiment was conducted with 36 participants who performed four tasks while data were collected using self-reports and objective physiological measures derived from electroencephalography (EEG). One of the most common method to estimate MWL is using an index/ratio of EEG band powers, known as Mental Workload Index (MWI). This study aimed to enhance MWL estimation by combining features from multiple domains, including statistical, spectral, and complexity-based measures, in contrast to relying solely on the MWI. Using various machine learning classifiers, we demonstrate that combining features from a diverse range of measures significantly improves the accuracy and robustness of MWL estimation. Multilayer perceptron demonstrated superior performance, with an accuracy of 94.5%. Furthermore, the inclusion of additional features beyond MWI leads to improved generalization across subjects and tasks, supporting our hypothesis that a multi-feature approach is more reliable for accurate MWL estimation. These findings provide a more robust solution for real-time MWL estimation, with potential applications in adaptive systems and user-centric design.