This paper presents a method for detecting cognitive load using electroencephalogram (EEG) signals, focusing on the extraction of features from intrinsic mode functions (IMFs) through variational mode decomposition (VMD). The study decomposes each EEG channel data into eight levels (4 seconds) using VMD, extracts entropy-based features from each IMF, and then classifies these features using supervised machine learning (ML) classifiers: LightGBM, XGBoost, and CatBoost. The experiments are conducted on two public EEG datasets, the multi-arithmetic tasks (MAT) and simultaneous task EEG workload (STEW). The performance is evaluated using metrics such as accuracy, specificity, sensitivity, positive predictive value, log-loss score, F1 score, and area under the receiver operating characteristic curve (AUROC). The proposed LightGBM classifier achieves superior classification accuracy rates of 97.22% and 95.51% for the MAT and STEW datasets, respectively, demonstrating enhanced precision in detecting cognitive load compared to existing methods. The study highlights the advantages of VMD in EEG signal decomposition and the effectiveness of ML classifiers in improving classification accuracy and sensitivity.This paper presents a method for detecting cognitive load using electroencephalogram (EEG) signals, focusing on the extraction of features from intrinsic mode functions (IMFs) through variational mode decomposition (VMD). The study decomposes each EEG channel data into eight levels (4 seconds) using VMD, extracts entropy-based features from each IMF, and then classifies these features using supervised machine learning (ML) classifiers: LightGBM, XGBoost, and CatBoost. The experiments are conducted on two public EEG datasets, the multi-arithmetic tasks (MAT) and simultaneous task EEG workload (STEW). The performance is evaluated using metrics such as accuracy, specificity, sensitivity, positive predictive value, log-loss score, F1 score, and area under the receiver operating characteristic curve (AUROC). The proposed LightGBM classifier achieves superior classification accuracy rates of 97.22% and 95.51% for the MAT and STEW datasets, respectively, demonstrating enhanced precision in detecting cognitive load compared to existing methods. The study highlights the advantages of VMD in EEG signal decomposition and the effectiveness of ML classifiers in improving classification accuracy and sensitivity.