The K-Nearest Neighbors (KNN) algorithm, effective for classification problems, is applied to improve routing in Benes optical networks. By integrating KNN with the network's routing table, this paper enhances the prediction of optimal paths. Experimental results show KNN prediction accuracy exceeding 64.47%, with general accuracy over 55%. Performance comparison among our proposed algorithm, conventional algorithm and manual selection is also presented in this paper. Our method significantly improves the extinction ratio (EXT) and the maximum measured bandwidth of the best paths, outperforming traditional methods. Additionally, the optimal paths predicted by KNN exhibit higher power at the receiving end and lower symbol error rates (SER) than the worst paths. Simulation results confirm the algorithm’s effectiveness in enhancing extinction ratio, bandwidth, and SER in Benes optical networks.