A Robust Fuzzy Time Series Forecasting Method Based on Multi-partition and Outlier Detection

We propose a robust fuzzy time series forecasting method based on multi-partition approach and outlier detection for forecasting market prices. The multipartition approach employs a specific partition criterion for each dimension of the time series. We use a Gaussian kernel version fuzzy C-means clustering to construct the fuzzy logic relationships and detect the outliers by calculating the grade of membership. We apply an additional model, which is trained on the set of outliers by Levenberg-Marquardt algorithm, for forecasting the outliers in testing set. The experiment results show that the proposed method improves the robustness and the average forecasting accuracy rate.