Maximum temperature forecasting in Giresun by using fuzzy time series methods

Time series forecasting is a critical problem for decision-makers in many fields. In this respect, it is widely studied in many fields such as finance, environment, natural sciences, health, engineering. Especially with the support of computer technology developed in the last few decades, the classical time series forecasting models, also known as probabilistic forecasting models, have been replaced by non-probabilistic models which are fuzzy logic-based and computational-based contemporary models. Among these modern time series forecasting models, fuzzy logic-based fuzzy time series forecasting models have an important place. When fuzzy time series forecasting models presented in the literature in time series forecasting problems are considered as a whole; Some of these studies appear to have been put forward to contribute to the fuzzification phase, some to identify fuzzy relationships, and some to contribute to the de-fuzzification phase and to advance forecasting performance. Also, with a different perspective; While some models use a first-order forecasting model, others aim to improve performance with high-order forecasting models. In this thesis, it is aimed to evaluate the results comparatively by forecasting the daily recorded maximum air temperature data of Giresun for the years 2006-2017 with different fuzzy time series forecasting models. In order to include the aforementioned two perspectives in the evaluation, first-order and high-order models using different approaches in the determination of fuzzy relationships have been used. Thus, the effects of both the model-order and the approaches used in determining fuzzy relations on performance were examined.Keywords: Fuzzy time series, maximum air temperature, forecasting, fuzzy relationships, model order