Yapay Sinir Ağları (YSA) ve ARIMA Modelleri ile Türkiye’de Aylık Sıfır km Otomobil Satış Adetlerinin Tahmin Edilmesi

Abstract

In this study, the estimation of brand new car sales in Turkey was carried out with Artificial Neural Networks (ANN) from artificial intelligence-based estimation methods and ARIMA Method (One of the Time Series Analysis techniques). In the study, the dependent variable is “monthly brand new car registrations”. The independent variables are; “Monthly automobile exports (USD), monthly automobile imports (USD), monthly American Dollar Rate (TL), monthly Turkey’s exports (USD) and monthly Turkey’s imports (USD)” values. Using the monthly data obtained from the Turkish Statistical Institute (TUIK) and the Central Bank of the Republic of Turkey (228 months-19 years, from the January 2002 to the end of December 2020), the sales of the number of brand new cars sold between January 2021 and March 2022 (15 month) were estimated by both methods.
 Artificial neural network architecture and various other parameters were determined using the data. By using the feed forward back propagation algorithm in ANN, the number of brand new automobile sales realized in the first three months of 2022 and the full year of 2021 has been estimated. Then, the number of automobile sales was estimated by using the ARIMA Method. The performance of the ANN was compared with the ARIMA Method according to the various indicators, and the estimation differences and results were interpreted. In the study, it was determined that ARIMA has gave better results compared to the ANN method. The reasons for the deviations in the estimations were interpreted.