Benchmarking Pattern Mining Algorithms for Service Non-Adherence: From Algorithms to Driver Insights

Cancellations and no-shows pose substantial challenges for service industries, yet research has primarily emphasized prediction rather than analysis of the underlying drivers, with most work centred on healthcare. This study presents an empirical comparison of four pattern mining paradigms—minimal correlated itemset mining, classical itemset mining, periodic pattern mining, and association rule mining—applied to non-adherence data. Using two real-world datasets from the beauty and hotel sectors, we evaluate algorithmic performance and the characteristics of Top-K patterns. The results illustrate differences in algorithm behaviour and pattern characteristics across datasets, with minimal correlated itemset mining and association rule mining providing interpretable and relevant patterns in this context. The findings further indicate that lead time and proximity to weekends are frequently associated with non-adherence in the analysed datasets, without implying generalizability beyond these cases.