A Novel ML-Driven Approach to Enhance CRN Performance under Varying Network Parameters

Abstract

This paper explores RL and DRL techniques for spectrum allocation in the context of CRNs, with consideration of difficulties like spectrum utilization and network performance in changing conditions. The proposed improved spectrum management model integrates RL with model-based prediction as a way of improving decision making. The results of the experiment prove that the identified approach allows for achieving an average level of accuracy of 96%, and a loss rate of 0.20, as well as of precision of 92% to 0.95. In addition, recall was extended from 0.85 to 0.90, and the F1 score was at 0.90, which indicated that the model demonstrated satisfactory performance at both precise and recall. The proposed algorithm outperformed existing machine learning models with a 96% accuracy, a low loss of 0.20, and an F1 Score of 0.90, showcasing superior reliability and adaptability. Based on these outcomes, it can be concluded that the proposed hybrid RL model is more effective in predicting the next available spectrum and more adaptable to the changes in the CRN environment than the single RL method thus, the proposed solution is suitable for real-time spectrum allocation in CRNs.