Making future predictions of Muş State Hospital patient radiological image numbers as a time series by using deep learning Methods

Time series forecasting processes are used in many areas of commerce, especially in science, health and engineering, and have a vital importance. Two different approaches are used in the time series estimation process. One of them is the statistics-based approach, and the second is the approach using machine learning models. Deep learning-based approach, which is a sub-branch of machine learning, is widely used in forecasting. Both approaches have their own strengths as well as weaknesses. Statistical based approaches such as ARIMA and Holt-Winters have been used for many years due to their robustness and flexibility. For the statistical-based approach, mediocre results can be obtained with the ARIMA, ETS, TBATS, THETAF and SES methods used in the R environment, even when the available data is limited. However, for deep learning approaches, insufficient data causes either the algorithm not working at all or very poor prediction results. However, it is stated in the literature that using deep learning and statistical-based approaches together as a hybrid leads to better results. In this study, different future predictions were made using the radiological image numbers of 11 years, in other words, 132 months, between the years 2010-2020, obtained from the Radiology unit of Muş State Hospital. In addition to deep learning algorithms such as MLP, NNTAR, ELM, statistical-based algorithms such as ARIMA, TBATS, HOLT-WINTERS, ETS, STL, THETAF and SES were also used for predictions. RMSE, MAE, MAPE and MASE metrics were used to evaluate the performance of the models. The data used in this study were handled as a time series in accordance with the explanation, modeling and estimation process. For this purpose, it is aimed to contribute to the literature by using a multi-model approach to estimate the monthly patient flow, taking into account the number of radiological images in different units of the hospital where the data are supplied. Two different general approaches, time series cross validation and forward estimation, were used in estimation processes and the results were evaluated. To put it more clearly, the research question investigated in this thesis is to prepare the relevant unit for the future by revealing to what extent the planning will be successful with the help of both statistical and deep learning prediction models for the future patient potential depending on the monthly historical data of the radiology unit. It is believed that the findings obtained will make important contributions to the future planning of the hospital by increasing both the service quality and patient satisfaction by facilitating the hospital managers in managing the patient flow to the hospital and referred to the radiology unit more efficiently.