Processing Online Massive Measuring Databases via Data-Uncertainty Quantifying Mechanism to Synthesize ANFIS

Abstract

Adaptive neuro-fuzzy inference systems (ANFISs) deriving from big data bring us the perspective in many fields. However, online performing both processing noisy and massive databases (NMDs) and training ANFISs is a challenge. Inspired by this aim, we propose a strategy with two phases, offline and online. The offline discovers an optimal data screening threshold (ODST) which is interpreted as an index to measure the uncertainty of the data in a data cluster. A new algorithm named A-ODST is proposed to estimate the ODST. Using the kernel fuzzy C-means clustering technique, a new filter named FbMU for blurring the ODST-based measured data-uncertainty is presented. An improved algorithm named NMD-ANFIS is presented to build the ANFIS from the NMD filtered by the FbMU. Based on the three main contributions of this paper to be the A-ODST, FbMU, and NMD-ANFIS, processing NMD and training ANFIS can be performed synchronously in the online phase. The combination of the solution to optimize the cluster data space via the NMD-ANFIS to simplify ANFIS’s structure and the filtering strategy of the FbMU for removing all the data points belonging to the data clusters with the highest uncertainty allows both filtering noise and reducing the size of the database to improve the calculating cost. Surveys from two experimental systems were carried out to verify these aspects. The compared results showed that the predicting error and the calculating time of the ANFISs built by the proposed method were better than that from the other surveyed methods.