<p>The increasing demand for high-speed data transmission and the difficulties caused by route selection and cross-talk (XT) limits are addressed in this research by introducing a unique deep learning-based method for spectrum allocation in optical networks. We proposed the Fragmentation Coefficient Core Changing (DL-FCCC) method, which optimizes spectrum distribution by using node mobility and dynamic core switching. The suggested approach first determines the shortest route using dijkstra algorithm and then it evaluates the fragmentation coefficient (FC), which is the largest contiguous block available for spectrum allocation, using the Continuous Aligned Slot Ratio (CASR) model. To guarantee that XT restrictions are adhered to FC values are arranged in ascending order and the smallest FC is chosen for spectrum allocation. This primary contribution of this work includes the dynamic spectrum allocation made possible by the DL-FCCC technology which minimizes XT interference and fragmentation to intelligently adjust to changing traffic demands. With notable decreases in the fragmentation ratio (FR), our approach leads to a 21.6% gain in spectrum utilization, with a significant reduction in fragmentation of 83.2% for USNET, 32% for NSFNET, and 16% for Indian networks respectively. Furthermore, the method significantly decreases the bandwidth blockage probability (BBP). At a traffic load of 1000 Gbps, the BBP is 0.19 for NSFNET, 0.09 for USNET, and 0.045 for an Indian network. We achieved SU value as 0.6,0.48,0.92 for NSFNET,USNET and indian network. CASR is 0.9994, 0.9922,0.9980 for NSFNET,USNET and indian network. This method incorporates machine learning for dynamic decision-making, improves energy efficiency, and is scalable for larger network infrastructures. Thorough testing using TensorFlow and Keras shows that the DL-FCCC strategy works better than conventional approaches, which makes it a viable option for improving overall network management effectiveness and optimizing optical network performance.</p>