Anxiety is a widespread psychological health issue, and binaural beats have emerged as a non-invasive treatment option. Electroencephalographic (EEG) signals help to track brain activity related to anxiety, but they often contain noise that affects accuracy. Traditional methods struggle with noise reduction and fail to capture the fine-grained temporal dynamics of EEG data, leading to poor classification accuracy. To address this, an Exponentially Magnificent Frigatebird Optimization-based CNN-LSTM (EMFO-CNN-LSTM) model is proposed in this research for classifying anxiety levels influenced by binaural beats. EEG data from the Binaural Beats on Anxiety Levels dataset is preprocessed using a Gaussian filter. Then, anxiety classification based on the binaural beats effect is done by Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM). The training of CNN-LSTM is done by Exponentially Magnificent Frigatebird Optimization (EMFO). The EMFO is designed by the combination of Exponentially Weighted Moving Average (EWMA) with Magnificent Frigatebird Optimization (MFO). The proposed EMFO-CNN-LSTM achieved 91.896% of accuracy, 91.989% of sensitivity and 91.448% of specificity.