at Google ScholarTitle data
Rebholz, Dominik:
Understanding Visitor Behavior: Causal Analysis of Touristic Time-Series Data Using Configurational Comparative Methods.
2025
Event: 1st European Symposium on Information Systems Engineering (ESISE)
, 11.09. - 13.09.2024
, Kempten, Germany.
(Conference item: Conference
,
Other
Presentation type)
Abstract in another language
Crowding poses significant challenges for managing popular tourist destinations. Nevertheless, the patterns at which crowding happens are often inadequately understood, posing challenges for effective crowd management. Traditional machine learning (ML) algorithms, while capable of predicting visitor numbers with reasonable accuracy, suffer from limitations such as high data requirements, overfitting, and a lack of transferability to different points of interest. Moreover, the complexity and black-box nature of these models necessitate the application of explainable AI methods to enhance interpretability and often require continuous computational resources. This study addresses these challenges by exploring the applicability of Configurational Comparative Methods (CCM), specifically Coincidence Analysis (CNA) and Qualitative Comparative Analysis (QCA), on time-series data to predict crowding. Unlike ML algorithms, CCM are adept at discerning causal relationships and do not solely rely on correlation-based analysis. Employing a dataset derived from camera sensors that track the number of people, bicycles, cars, and other vehicles at a lakeside touristic spot, we aim to develop a causal model that identifies the relevant feature combinations (e.g., high temperature and day of the week) contributing to crowding. Our findings demonstrate that CCM can successfully be applied to time-series data, revealing actionable insights into the interaction of various conditions leading to crowding. This causal model offers an interpretable and transferable framework for managing visitor behavior at diverse touristic sites, thus providing a robust alternative to traditional ML approaches.