Minimum weather temperature forecasting of giresun province using long short-term memory deep artificial neural network

In recent years, it has been observed that deep learning methods give better results in time series forecasting compared to classical time series methods, fuzzy time series methods and other artificial neural network methods. With the development of deep learning, long short-term memory, which is one of the innovative network models formed in artificial neural networks and is also a deep learning method, has started to be used in time series analysis in the literature. As in artificial neural networks, the initial values of the model created in deep learning methods and the weight values in the model affect the time series forecasting performance of the model. It is important to determine these values for forecasting accuracy and they appear as problems to be solved in the literature. Within the scope of this thesis, the single layered long short-term memory approach, using the Xavier method in generating the initial values of the model and the Adam algorithm for optimizing the layer weights, was utilized. The daily average minimum temperature real-world time series in Giresun province was used to show the forecasting performance of the approach. The approach was compared with some other forecasting methods in the literature in terms of forecasting performance and it was found that the approach has a superior forecasting performance compared to other methods.