Power Allocation RBF-type Neural Network Regression Method

A novel method, the Power Allocation Radial Basis Function (PA-RBF), is developed in this study. It is based on Decision Tree (DT) classification and RBF neural network regression. In PA-RBF, the DT algorithm is applied to separate the whole dataset into segments, after which smaller RBF models are applied on the separated datasets. Then trial and error method is utilized to find the optimum Division Percent (DP) and Minimum Parent Size (MPS) characteristics of the DT algorithm, as well as the RBF models’ number of hidden neurons and spread amounts. The regression performance of the RBF and PA-RBF methods is compared by using two different case studies of open channel junction velocity prediction and Parkinson’s disease tracking. The results show that in the case of open channel junction velocity simulation, the PA-RBF model with RMSE of 0.142 performs about 90% more accurately than the RBF model with RMSE of 0.27; as for Parkinson’s disease tracking, PA -RBF with RMSE of 7.14 shows nearly 22% more accurately compared with the RBF model with RMSE of 8.75.

کلیدواژه ها:

Classification ، Decision tree ، Power allocation ، Radial basis function ، Regression.