Forecasting of COVID-19 confirmed and death cases in Turkey using arima and deep learning models

The coronavirus (Covid-19) is a major worldwide threat that emerged in 2019, spread rapidly, and affects human health. It is thought to originate from seafood and animal markets of Wuhan city, China. Covid-19 is a contagious disease and is transmitted from person to person. On March 11, 2020, the World Health Organization (WHO) declared the Covid-19 outbreak a global pandemic. It is of great importance for countries to be able to forecast the number of confirmed and death cases caused by the epidemic so that they can plan for the future. The aim of this thesis work is to forecast Covid-19 total confirmed and death cases of Turkey based on previous data. The dataset was created using the data on the website of the Ministry of Health of the Republic of Turkey. Three different deep learning models as Long Short Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), and Autoregressive Integrated Moving Average (ARIMA) statistical model was used as prediction models. Data from March 11, 2020, to May 31, 2021, were used to train and test the models, and the number of confirmed and death cases between June 1 and June 30, 2021, was forecasted. Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were used to evaluate the performance of the models. The ARIMA model outperformed deep learning techniques in terms of RMSE, MAPE and forecasting values. The forecasted values produced by the ARIMA model are more compatible with the actual number of confirmed and death cases in Turkey. Keywords: Autoregressive Integrated Moving Average, Bidirectional LSTM, Covid-19, Deep Learning, Gated Recurrent Unit, Long Short Term Memory