Hybrid Machine Learning Methodology for Real Time Quality of Service Prediction and Ideal Spectrum Selection in CRNs

The application of wireless communication is very complex and there is always a demand for accurate Quality of Service (QoS) for estimating and optimize the spectrum allocation in Cognitive Radio Networks (CRNs). Current machine learning models frequently struggle to adapt effectively to change the network conditions due to significant computational complexity and constrained real-time performance. This paper presents a Hybrid Deep Learning and Ensemble Regression Model (HyDERM) to address these limitations in real-time QoS prediction and spectrum decision-making. The proposed HyDERM model integrates Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Networks (ANN) to enhance accuracy and effectiveness. Key metrics such as Signal-to-Noise Ratio (SNR), bandwidth availability, network load, latency, packet loss, and interference level are evaluated for QoS assessment. The model is assessed using five advanced machine learning techniques: Polynomial Regression, SVR, RF, Gradient Boosting, and ANN. The results demonstrate that HyDERM achieves a R² value of 0.96, exceeding all the compared models. It reduces Mean Squared Error (MSE) by 23% and Mean Absolute Error (MAE) by 19%, illustrating its effectiveness. The results show that the suggested HyDERM can improve frequency efficiency and allow for smooth communication, making it a feasible choice for the next generation of wireless networks.