Average ticket prices of airlines have been falling continuously in recent years. This is true not only for our country, but also for Europe and America. Moreover, this situation is observed both in international and domestic lines. Since ticket revenues are the most important source of income for airlines, the fact that they are constantly decreasing due to factors such as increasing competition increases the importance of determining ticket prices. It turns out that systematic approaches in determining ticket prices are weak and human judgment is an important factor. In addition, many factors affect the determination of ticket prices. Therefore, systematic approaches that are tried to be developed can produce low statistical performance values.
In this study, the factors affecting the determination of ticket prices were investigated and a Decision Support System model that could be used for Revenue Management was produced. Contrary to the fact that many studies and products in the literature aim to increase income, this study aims to increase profits. Airline expenses were researched and the characteristics of target revenue and available seat kilometer cost (CASK) value were included in the research for the first time. In this study, the actual values of an airline operating both domestic and international scheduled and charter flights in Turkey were used. In order for the model to learn from the data that produced the best profit, the data that produced the best profit performance from the obtained data were separated and used in model training. In order to increase the data quality, outlier data were removed and the data obtained at each stage were collected in different data sets and analyzed separately. Trendline data was also created in order to predict the negative effects of price fluctuations. All these studies were applied to six different models besides one developed model and tested according to six different statistical performance evaluation criteria.
In this study, the data working on it were evaluated as overlapping windows, not as a continuous time series. A new model based on variable batch size hyperparameter, not static, was developed and used on these datasets. This new approach on the datasets and the new model developed produced more successful performance criterion values than other models.
With the proposed model, a Decision Support System model was created that produces high performance criteria in which human judgment is reduced. This model can be used in the airline industry as well as in other industries with overlapping window data structures. |