Bayesian Ridge Regression Integrated Algorithm Based On Stochastic Configuration Networks

With the increasing maturity of machine learning algorithms,people have more experience in processing various types of complex data.Enabling machine learning algorithms to better handle outliers in complex data has become an important topic in the field of machine learning algorithms.In recent years,Bayesian inference has solved various difficult problems in classical statistics and has been successfully applied in many fields such as actuarial calculations,quality control and evaluation,and engineering reliability evaluation.Bayesian inference uses new data to continuously update the cognition of unknown parameters through repeated iterations.This feature makes Bayesian inference have a strong ability to adapt to the data,and can fully learn samples when training the model Data and have a deeper understanding of unknown parameters,which also make Bayesian inference have more robust performance when dealing with data containing a large number of outliers.Stochastic configuration network is an important innovation of a random weight neural network.It avoids the shortcomings that traditional neural networks can not solve and also solves the shortcomings of the Random Vector Functional Link(RVFL)Network in the selection of weight parameters.However,the least square method used by the Stochastic Configuration Networks when estimating the output weight makes the Stochastic Configuration Networks perform poorly in the face of noisy data,and the influence of outliers will be more obvious.In response to this problem,this paper proposes a Bayesian Ridge Regression Algorithm(BSCN)based on Stochastic Configuration Networks.First,the Stochastic Configuration Networks(SCN)uses an effective supervision mechanism to select the input weights and deviations of the network,and gradually increases the number of hidden layer neurons according to the residual of the model,ensuring the algorithm's excellent generalization ability and stable network performance,Which also optimizes the number of random neural network nodes.The Bayesian inference is applied to the Stochastic Configuration Networks,which optimizes the estimation method of output weights,reduces the influence of the algorithm from outliers,and effectively improves the robustness of the algorithm while retaining the advantages of the random configuration network.On this basis,BSCN is applied in the framework of an integrated regression learning algorithm,and a Bayesian Ridge regression integrated algorithm based on Stochastic Configuration Networks(Bagging-BSCN)is proposed,which further improves the generalization ability and stability of the algorithm.The comparison of experimental results on simulation data and standard data sets shows that the Bayesian Ridge Regression Integrated Algorithm(Bagging-BSCN)based on stochastic Configuration Networks perform better,which is compared with a stochastic configuration networks algorithm and another random weight neural networks in the face of noise data.