Forecasting Better Prices for Trip Packages based on Historical Sales Data and Related Factors: In the context of Europe Railway Tourism

Abstract

Tourism is a valuable income source for any country. Having a competitive price is crucial for surviving the current environment. This study is on forecasting better prices for a railway tourism company in Europe, considering the past sales patterns with external factors. Currently, they are deciding prices manually rather than using a scientific method. But manual price allocation is not reliable, as it is subjective. To mitigate this issue researchers tried to implement a scientific method for price forecasting. In this study, the performance of Deep Neural Network (DNN), Ordinary Least Squares (OLS) Multiple Linear Regression (MLR) Model, SARIMAX Model, Support Vector Machine, and Extreme Learning Machine was evaluated; which were used by past researchers for price prediction. The acquired dataset contained 89 trip packages and the DNN had the least root mean square error for 75 packages and OLS MLR was the best for the other 14; 11 and 3 using statsmodels and sklearn libraries respectively. As a single model could not be selected as the best model, a hybrid model was created. The hybrid model contained a DNN, an OLS MLR model, a Linear Interpolation function, and a revenue-maximizing function. The theoretically estimated increase in revenue for the hybrid model had a maximum, minimum, and average of 120.59%, 12.12%, and 79.25% respectively. It was concluded that the DNN and OLS MLR models perform best when predicting prices while linear interpolation performs best for interpolating trip prices.