<p>Attention has a significant impact on cognitive abilities including learning and deciding. Under the fast growth of internet technology, virtual reality has also greatly progressed. Exploring the impact of virtual reality games on attention is of great importance, but a quantitative analysis of the influence of virtual games on attention is inadequate. To address this issue, a method based on support vector machines is invented to judge and classify electroencephalography signals. Firstly, a method based on wavelet thresholding and ensemble empirical mode decomposition is proposed to address the problem of difficult extraction and processing of electroencephalography signals. This method can preprocess the real-time electroencephalography signals collected from users in different attention states, and then classify the signals using an improved support vector machine. Experiment results showed that when the training set size was 500, the root mean square error values of the wavelet thresholding method, ensemble empirical mode decomposition, and hybrid algorithm were 0.25, 0.19, and 0.16. The values of the hybrid algorithm for the alpha, beta, theta, and delta frequency bands were 0.09, 0.15, 0.19, and 0.11, respectively. When the dataset size was 800, the root mean square error values of the convolutional neural network model, support vector machines model, and improved support vector machines model were 0.19, 0.16, and 0.13, respectively. The research results indicate that the proposed hybrid denoising algorithm and improved support vector machines model have good model performance, providing theoretical support for attention training and testing in fields such as medicine and education.</p>