Multi-Layer Perceptron (MLP) Neural Networks for Time Series Classification
Arash Gharehbaghi · 2023
Multi-layer perceptron neural networks have been widely used for different applications of time series analysis, especially for the classification purposes such as speech recognition over decades. Architecture of a MLP does not allow the classifier to learn dynamics of a time series, and therefore one should find a way to preserve the dynamics of a time series at the feature level, or in another word at the input node of the MLP. To this end, Time-Delayed Neural Network was introduced based on the MLP architecture for a problem of speech recognition, which is introduced in a section of this chapter. Time-Growing Neural Network was later introduced as an elaborated version of the former one in which the learning process is further improved. Time-Growing neural network, along with the different schemes of the growing windows for the implementation are discussed in this chapter. This chapter is finished by describing privileges of Time-Growing neural Network over the other alternatives in the theoretical manner.