Robust learning algorithm based on particle swarm optimization for dendritic artificial neural networks in time series forecasting

Time series forecasting is a crucial problem encountered in a variety of fields. Although there are a great number of methods in the literature, they can be reviewed under two basic title as probabilistic and non-probabilistic methods. Especially, as computational-based time series forecasting methods, different kinds of artificial neural networks (ANNs) have been widely and successfully used in the literature. While some of them use additive aggregation function, some of them use multiplicative aggregation function in the structure of their neuron models. Dendritic Neural Networks, proposed in recent years, have also both additional and multiplicative neuron models, together. It is inevitable that the forecasting performance of such an artificial neural network will be negatively affected by the outliers that the time series of interest may contain due to the neuron model in its structure. In this thesis, a robust learning algorithm is proposed for dendritic artificial neural network training for the first time in the literature. The proposed robust learning algorithm uses Huber's loss function as a fitness function. The iterative process of the robust learning algorithm is carried out by particle swarm optimization, an artificial intelligence optimization algorithm. The performance of the dendritic artificial neural network trained with the proposed robust learning algorithm has been demonstrated by the analysis of different real-life time series and the analysis of the contaminated time series obtained by injecting different numbers and scales of outliers. The obtained results show that the dendritic artificial neural network trained by the proposed robust learning algorithm produces satisfactory predictive results in the analysis of time series with and without outliers.