Ski-day patterns from RFID lift usage data

Abstract

RFID-based access systems generate detailed records of skier lift usage, enabling large-scale analysis of skiing behaviour but posing challenges due to the sequential and unlabelled nature of the data. In this study, we analyse skier movement patterns on data collected at Mt. Kopaonik over six ski seasons, comprising more than 7.6 million gate passages across 22 ski lifts. Each skier day is modelled as a sequence of lift usages enriched with temporal information. We learn compact behavioural representations using a self-supervised Transformer-based contrastive learning framework that captures both spatial and temporal structure. The learned embeddings are clustered using k-means. The resulting ten clusters reveal distinct but overlapping skiing behaviours, including repetitive looping within central lift areas, extended exploratory ski days spanning multiple resort zones, short-duration sessions, and structured progression across elevations. This paper thus introduces a method for modeling complex spatio-temporal event data, and transformation to a representation where traditional analytics can be performed; and is also applied to a real-world dataset to produce descriptive analysis of skiing patterns.

Sandro Radovanović

Assistant Professor at University of Belgrade

My research interests include machine learning, development and design of decision support systems, decision theory, and fairness and justice concepts in algorithmic decision making.