Graph kernels are methods used in machine learning algorithms for handling graph-structured data. They are widely used for graph classification in various domains and are particularly valued for their accuracy. However, most existing graph kernels are not fast enough. To address this issue, we propose a new graph kernel based on the concept of entropy. Our method has the advantage of handling labeled and attributed graphs while significantly reducing computation time when compared to other graph kernels. We evaluated our method on several datasets and compared it with various state-of-the-art techniques. The results show a clear improvement in the performance of the initial method. Furthermore, our findings rank among the best in terms of classification accuracy and computation speed compared to other graph kernels.",