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

Abstract

Abstract Traveling and tourism is a world-leading industry. Having a competitive price is crucial to survive in the current competitive environment.This study is related to forecasting better prices for a railway based tourism company located in the European region, by considering the past sales patterns with the external factors such as weather, season and holidays. Currently, they are deciding their prices by past experiences. The dataset received contained all products of flam railway and by preprocessing and feature extraction relevant records were chosen only relating to flam railway. Extracted data was combined with external data such as weather,season and holidays which were collected from APIs. Then trip packages were identified and dataset was broken based on different packages. In this research researchers try to evaluate and compare the performance of various models traditionally used for price prediction. Here performance of Deep Neural Network, Ordinary Least Squares Multiple Linear Regression Model (sklearn), SARIMAX Model,Ordinary Least Squares Multiple Linear Regression Model (statsmodels), Support Vector Machine and Extreme Learning Machine was evaluated. Root mean square error was used to compare the performance of the models. Model with least root mean square error was selected to predict the price of the model. 75 packaged had Deep Neural Network as the best performing model while 11 and 3 packages had Ordinary Least Squares Multiple Regression (statsmodels) Model and Linear Regression Model(sklearn) respectively. The hybrid model was created using above models, deflated price and respective sales volume. Estimated increase of revenue when using the better price of the hybrid model had a maximum of 120.59% increase, minimum of 12.12% increase and average of 79.25%