A Novel Median Dendritic Neuron Model for Prediction
Shi Wang, Yang Yu, Lin Zou, Sheng Li, Hang Yu, Yuki Todo, Shangce Gao · IEEE Access · 2020
Dendritic neuron model (DNM) that utilizes a single dendritic neuron to emulate the information processing in human brains has been successfully applied to classification, approximation, and prediction fields. However, it is still a challenging task to improve the performance for forecasting the time series with outliers. A time series is a collection of data ordered in time. An outlier in time series as an input can affect prediction performance. Thus, in this study, we first propose a novel median dendritic neuron model (MDNM) for prediction to deal with time series with outliers. In MDNM, DNM and median function are combined to achieve better forecasting results. Moreover, states of matter search (SMS) algorithm is used to train MDNM. Twenty real time series and a benchmark time series are adopted to verify the performance of MDNM. The comparison between the MDNM and DNM suggests that MDNM is effective for forecasting time series with outliers.