Similarity and wavelet transform based data partitioning and parameter learning for fuzzy neural network

Abstract

Function approximation is an important task in many different fields like economics, engineering, computing, classification, and forecasting. From a finite data set, the basic task of a function approximation method is to find a suitable relationship between variables and their corresponding responses. In the recent past, the improved neural networks including intuitive, interpretable correlated-contours fuzzy rules for classification tasks were proposed. However, the acquired data set can contain large volume of data and noise that degrades the classification ability of the model and increases the computational time. Thus, it is important to consider this problem which was not focused on recent existing works. Furthermore, there are also some neuron regularization issues in the second layer. To solve this issue in this proposed system Bat optimization based feature selection is proposed for optimal selection of features from the available dataset. Then classification is done by using enhanced neural network including intuitive and interpretable correlated-contours fuzzy rules (EC-FR). According to fuzzy rules extraction, an appropriate framework is built-in which similarity-based directional component of data partitioning and also a model to form cloud data is presented. Neurons weight and bias values are computed by adapting wavelet functions. Finally, parameters of the fuzzy neural networks are fine-tuned using the hybrid ant colony particle swarm optimization (HASO). Performance is evaluated primarily in accordance with the subsequent metrics like precision, recall, accuracy, and error rate.