Predictive Analysis of Machine Learning Error Classification Based on Bayesian Network

Abstract

With the emergence of various types of wireless communication products and the explosive growth of wireless network services, spectrum resources have become increasingly scarce. As the basis of cognitive radio technology, cognitive devices need to have real-time spectrum sensing capabilities, and classify and predict the perceived radio signals. Introducing Bayesian network for model construction, and using genetic algorithm to optimize and improve the model structure can more accurately establish the optimal network directed graph in structural learning, so as to better explore the high-risk predictive factors of machine learning. This paper makes a comparative analysis of the basic methods of the Bayesian network model in structural learning and parameter learning, and discusses the algorithm models that may be used to improve the structural learning. The adaptive search scoring function genetic algorithm is introduced into the network topology map construction part of the Bayesian network model structure learning, and a Bayesian network machine learning prediction model based on genetic optimization is proposed. The prediction algorithm of the maximum posterior probability of the perceptual signal considers which system the signal is most likely to come from at the next moment, and uses the estimation of the signal characteristics of the system as the signal prediction result. This paper further derives the upper bound of signal feature estimation error. Based on the prediction algorithm, the cognitive device can avoid the communication channel that will be occupied by the main user system in advance, thereby saving channel switching overhead. Simulation analysis verifies that the prediction algorithm designed in this paper has better prediction accuracy, and shows that as the number of signal samples sensed increases, the accuracy of the prediction algorithm is also improved accordingly. The classification accuracy rate based on Bayesian network is above 83% and the fluctuation rate is reduced to 2.5%.