Robust Fuzzy Inference System for Prediction of Time Series with Outliers

In this paper, a new robust fuzzy inference system is utilized to predict the chaotic time series with noises or outliers. We employ an improved fuzzy rule extraction algorithm using data mining concepts to make the resulting fuzzy system more robust with respect to the input noises or outliers. And the fuzzy inference system is optimized with a partition refinement strategy so that a more suitable topology is determined by the training data. The proposed techniques in this paper are examined, with comprehensive robustness analysis, by a classical benchmark time series forecasting problem and a real world application of ship zig-zag test. The results and comparisons show that our method performs favorably in terms of both accuracy and robustness.