1715 KB35 KB198520262025Kumar, Chellappan AgeesSindhu, Ayya Dhurai Suceelalštevilka:2letnik:55str. 77-86ISSN:0352-9045DOI:10.33180/InfMIDEM2025.201COBISSID_HOST:281450499URN:URN:NBN:SI:doc-GIGFZIUQenStrokovno društvo za mikroelektroniko, elektronske sestavne dele in materialeInformacije MIDEMCNNCoati optimizacijaCoati optimizationdeep learninggated dilatedGated Dilated CNNGhostNetgloboko učenjenetwork slicingrezanje omrežjaCoati optimized hybrid neural network for efficient network slicing in 5 generation network| Coatijevo optimizirano hibridno nevronsko omrežje za učinkovito rezanje omrežja v omrežju petih generacij|Network slicing (NS) divides the physical network into many logical networks in order to support the variety of new applications with higher performance and flexibility needs. As a result of these applications, a massive amount of data has been generated with a huge number of mobile phones. Due to this, NS performance has been greatly impacted and extreme challenges have been created. To efficiently handle the challenges, this paper proposes a novel Optimal Network slice Classification Using Deep learning (ONE-CLOUD) technique, which integrates the Coati Optimization Algorithm (COA), GhostNet, and Gated Dilated Convolutional Neural Network (CNN). COA optimizes features such as user device type, packet loss ratio, and delay rate, employing GhostNet model, and Gated Dilated CNN for network slice classification. The proposed method classifies network slices into enhanced Mobile BroadBand (eMBB), Ultra-Reliable and Low-Latency Communications (URLLC), and massive Machine-Type Communications (mMTC). The effectiveness of the suggested approach has been evaluated using the 5G-SliciNdd dataset, utilizing evaluation criteria like accuracy, precision, recall, sensitivity, specificity, throughput, and reduced latency. The overall accuracy of the proposed method is 5.78%, 2.78% and 4.70% higher than the existing DQN-E2E, DRL, and AAA techniques respectivelyRazrez omrežja (NS) razdeli fizično omrežje na več logičnih omrežij, da bi podprl različne nove aplikacije z večjo zmogljivostjo in prilagodljivostjo. Zaradi teh aplikacij se je z velikim številom mobilnih telefonov ustvarila ogromna količina podatkov. To je močno vplivalo na zmogljivost omrežja NS in povzročilo izjemne izzive. Za učinkovito obvladovanje teh izzivov članek predlaga novo tehniko optimalne klasifikacije omrežnih rezin z uporabo globokega učenja (ONE-CLOUD), ki združuje algoritem COA (Coati Optimization Algorithm), GhostNet in gated dilated konvolucijsko nevronsko mrežo (CNN). COA optimizira lastnosti, kot so vrsta uporabniške naprave, stopnja izgube paketov in stopnja zamude, pri čemer uporablja model GhostNet in Gated Dilated CNN za klasifikacijo omrežnih rezin. Predlagana metoda razvršča omrežne rezine v izboljšano mobilno širokopasovno omrežje (eMBB), izjemno zanesljive komunikacije z nizko zakasnitvijo (URLLC) in množične komunikacije strojnega tipa (mMTC). Učinkovitost predlaganega pristopa je bila ocenjena z uporabo podatkovne zbirke 5G-SliciNdd, pri čemer so bila uporabljena merila za ocenjevanje, kot so natančnost, točnost, priklic, občutljivost, specifičnost, prepustnost in zmanjšana zakasnitev. Skupna natančnost predlagane metode je za 5,78 %, 2,78 % in 4,70 % višja od obstoječih tehnik DQN-E2E, DRL in AAATEXTznanstveno časopisjejournalsSlovenian National E-content AggregatorNational and University Library of SloveniaStrokovno društvo za mikroelektroniko, elektronske sestavne dele in materiale