This study aims to classify cognitive workload levels from EEG signals. EEG signals from 48 subjects under resting and task cognitive load conditions were analyzed. Noise and artifacts were removed by applying band-pass and notch filtering methods in the 1-50 Hz band on the EEG data. Then, the EEG data were segmented with the windowing technique in 256 and 512 sample sizes, and a total of 309 features based on time, frequency, and complexity were extracted. Using the obtained feature set, logistic regression (LR), support vector machines (SVM), k-nearest neighbor (k-NN), random forest (RF), XGBoost machine learning (ML) algorithms and deep neural networks (DNN), one-dimensional convolutional neural networks (1D-CNN) and long short-term memory (LSTM) deep learning (DL) methods were applied for multi-class classification. In the experimental results, the highest success was obtained in the XGBoost model with a 99.4% accuracy rate and 0.990 Cohen’s kappa value, and in DL methods, a 98.75% accuracy rate and 0.981 Kappa value in the LSTM model. This study reveals that integrating multidimensional features obtained from EEG signals with both ML algorithms and DL models provides high accuracy in cognitive workload classification.