Classification analysis with fuzzy wavelet kernel extreme learning machines approach

Classification problems are one of the problems frequently encountered in different data types in daily life. Recently, various machine learning algorithms have been proposed to solve such problems, and these algorithms have some unique advantages. The time-frequency localization properties of wavelets, the learning capabilities of the artificial neural network, the approximate reasoning feature of the fuzzy inference system, and the advantages of the extreme learning machine algorithm such as one-pass learning at an extremely high learning speed and good generalization performance can present an effective solution for classification problems. In this context, this thesis combines these advantages and presents a new structure called fuzzy wavelet kernel-based extreme learning machine (FWKELM). The main goals of FWKELM are to significantly reduce network complexity by reducing the number of linear learning parameters and maintain acceptable accuracy and generalization performances. In the proposed structure, each fuzzy rule corresponds to a sub-wavelet neural network and consists of wavelets with different dilation and translation. Within the scope of the study, firstly, it is shown that a Mexican Hat wavelet is an acceptable extreme learning machine algorithm kernel, and then its equivalence to a kernel extreme learning machine is proven with a fuzzy model. To evaluate the classification accuracy of FWKELM, it is compared with various popular methods on different type classification problems. The simulation results show that the proposed approach is more robust and has remarkable efficiency.