This study presents an EEG-based method for detecting cognitive load using variational mode decomposition (VMD) and a LightGBM classifier. The method involves decomposing EEG signals into eight intrinsic mode functions (IMFs) using VMD, then extracting entropy-based features from each IMF. These features are used to classify cognitive load using supervised machine learning classifiers, including LightGBM, XGBoost, and CatBoost. The method was tested on two public EEG datasets: the Multi-Arithmetic Tasks (MAT) dataset and the Simultaneous Task EEG Workload (STEW) dataset. The results showed that the LightGBM classifier achieved high classification accuracy rates of 97.22% and 95.51% for the MAT and STEW datasets, respectively. The study demonstrates that the proposed technique can detect cognitive load more accurately than existing methods. The use of machine learning and data mining techniques improved the accuracy and sensitivity of predicting cognitive load. The study highlights the importance of VMD in EEG signal decomposition and analysis, as it allows for the preservation of EEG signal energy and information, and adapts to the features of EEG data and research objectives. The study also shows that the use of conventional decomposition techniques and machine learning classifiers can lead to lower accuracy levels in cognitive load or stress detection. The main contributions of this study include the use of VMD for EEG signal decomposition, the extraction of entropy-based features from different decomposition levels, and the application of XGBoost, LightGBM, and CatBoost for classification.This study presents an EEG-based method for detecting cognitive load using variational mode decomposition (VMD) and a LightGBM classifier. The method involves decomposing EEG signals into eight intrinsic mode functions (IMFs) using VMD, then extracting entropy-based features from each IMF. These features are used to classify cognitive load using supervised machine learning classifiers, including LightGBM, XGBoost, and CatBoost. The method was tested on two public EEG datasets: the Multi-Arithmetic Tasks (MAT) dataset and the Simultaneous Task EEG Workload (STEW) dataset. The results showed that the LightGBM classifier achieved high classification accuracy rates of 97.22% and 95.51% for the MAT and STEW datasets, respectively. The study demonstrates that the proposed technique can detect cognitive load more accurately than existing methods. The use of machine learning and data mining techniques improved the accuracy and sensitivity of predicting cognitive load. The study highlights the importance of VMD in EEG signal decomposition and analysis, as it allows for the preservation of EEG signal energy and information, and adapts to the features of EEG data and research objectives. The study also shows that the use of conventional decomposition techniques and machine learning classifiers can lead to lower accuracy levels in cognitive load or stress detection. The main contributions of this study include the use of VMD for EEG signal decomposition, the extraction of entropy-based features from different decomposition levels, and the application of XGBoost, LightGBM, and CatBoost for classification.