Enhancing Mental Workload Prediction through LightGBM during Multitasking

Enhancing Mental Workload Prediction through LightGBM during Multitasking

Stať ve sborníku mimo WoS a Scopus

Multitasking is an essential aspect of daily life; however, it significantly increases mental workload (MWL), which can affect cognitive performance, decision making, and overall effectiveness. Thus, accurately assessing MWL is significant in various fields, including human-computer interaction, aviation, and healthcare, where cognitive overload can lead to unsuitable decisions. The brain computer interface (BCI) based on electroencephalography (EEG) presents a viable, non-invasive option for real-time monitoring of MWL, allowing an adaptive system to improve performance and user experience. However, because EEG patterns vary widely among individuals, it is still challenging to develop a generalized MWL prediction model. Therefore, Light Gradient Boosting Machine (LightGBM) with manually extracted features is proposed. Our analysis was based on the "STEW" dataset, which includes two task conditions: "No task" and a multitasking activity using the SIMKAP framework. The proposed model achieved an average accuracy of 84.0% (±14.4%) and an average F1-score of 83.1% (±18.2%), showcasing its strong predictive performance while maintaining computational efficiency compared to deep learning methods. These results highlight LightGBM’s potential as a fast, subject-independent MWL classification tool, therefore enabling the design of scalable and flexible cognitive monitoring systems for practical use.

Multitasking is an essential aspect of daily life; however, it significantly increases mental workload (MWL), which can affect cognitive performance, decision making, and overall effectiveness. Thus, accurately assessing MWL is significant in various fields, including human-computer interaction, aviation, and healthcare, where cognitive overload can lead to unsuitable decisions. The brain computer interface (BCI) based on electroencephalography (EEG) presents a viable, non-invasive option for real-time monitoring of MWL, allowing an adaptive system to improve performance and user experience. However, because EEG patterns vary widely among individuals, it is still challenging to develop a generalized MWL prediction model. Therefore, Light Gradient Boosting Machine (LightGBM) with manually extracted features is proposed. Our analysis was based on the "STEW" dataset, which includes two task conditions: "No task" and a multitasking activity using the SIMKAP framework. The proposed model achieved an average accuracy of 84.0% (±14.4%) and an average F1-score of 83.1% (±18.2%), showcasing its strong predictive performance while maintaining computational efficiency compared to deep learning methods. These results highlight LightGBM’s potential as a fast, subject-independent MWL classification tool, therefore enabling the design of scalable and flexible cognitive monitoring systems for practical use.

mental Workload, Prediction, EEG, Brain Computer Interface (BCI)

mental Workload, Prediction, EEG, Brain Computer Interface (BCI)

AZHAR ALI, S.; AL-QURAISHI, M.; EL FERIK, S.; MALIK, A.

15.07.2025

IEEE

Croatia

979-8-3315-0338-3

2025 11th International Conference on Control, Decision and Information Technologies (CoDIT)

1267

1271

5

https://ieeexplore.ieee.org/document/11321684