Time series prediction with a new convolutional neural network model

Time series forecasting has always been an attractive issue in the most scientific area. In particular financial time series forecasting has become quite attractive topics owing to its broad implementation areas and substantial impact. Therefore, various types of computational intelligence techniques such as convolutional neural networks (CNNs) have been used for financial time series forecasting. However, in these studies reported so far, CNNs have been rarely preferred for the forecasting of financial time series. And almost all-studies time sequence effect of time series is not preserved on forecasts because of image transformation. From this point of view, within the scope of this thesis, by introducing a new CNN prediction model, it is aimed to prevent information loss that may occur in the image transformation process and to successfully reveal the properties of time series. The proposed CNN forecasting model consists of three convolutional layers and five fully connected layers, also Relu and Elu activation functions have been preferred to determine the nonlinear relations between inputs and outputs. In order to evaluate the performance of the proposed system, the model has been applied to Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and London stock market data, which are frequently used in financial time series literature. The results have been evaluated on different aspects as an error criterion, a regression analyses and also a visual demonstration. Outstanding forecasting performance of the proposed CNN model has been observed by comparing the obtained results with some state-of-the-art forecasting tools such as different kinds of Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), fuzzybased approaches, and some traditional methods.