Predicting Passenger Demand and Optimizing Fleet Allocation: A Machine Learning Approach for Icelandic Tour Operators

Abstract

The need for more efficient resource allocation methods in tour operations has been created by the growing complexity driven by increasing tourism in destinations like Iceland. In this study, the application of machine learning techniques to optimize tour operations through improved passenger forecasting and bus assignment is explored. Three key questions are addressed: Can a system be constructed to improve bus assignment in tour operations? How are passenger numbers and forecasting accuracy impacted by external factors? How effectively can forecasting inaccuracies be compensated for by intelligent bus assignment algorithms?A system was developed by integrating internal tour operator data, weather information, and tourist arrival statistics. Four forecasting models were implemented and compared: Random Forest, Convolutional Neural Network, Support Vector Regression, and Long Short-Term Memory (LSTM) networks. An intelligent bus assignment system using a Best Fit greedy algorithm was developed to utilize these forecasts.The best performance was demonstrated by the LSTM model, with a Mean Absolute Error of 9.08 passengers being achieved. At a 100% utilization target, a 22.76% overbooking rate was shown by the system, indicating that 77.24% of tours could be handled without additional monitoring or manual intervention. This rate was further improved when the utilization target was adjusted to 90%, resulting in only 13.77% of tours requiring manual oversight. A 74.41% utilization rate was achieved by the bus assignment system, comparable to the 79.54% rate of manual methods, while significantly reducing the need for manual adjustments.A contribution to the field of machine learning applications in tourism management is made by this study through the demonstration of the potential for data-driven approaches to enhance operational efficiency. While the proposed system shows promise in reducing manual workload, areas for future research are also highlighted, particularly in improving forecasting accuracy and managing overbooking risks in dynamic tourism environments.