Abstract: High-utility co-location patterns refer to those with strong instructive value and application-driven potential. Its extraction is based on utility values associated with dataset features or with individual instances. Research indicates that the distance
distribution between feature-participating instances and neighboring instances, as well as the co-located instances formed by
participating instances, reflect the utility of feature-participating instances in the pattern. Therefore, this study defines metrics for instance participation based on row instance utility, instance participation pattern utility, feature average participation utility, and pattern average participation utility, and proposes an algorithm for mining high average-utility co-location patterns based on instance distribution. The algorithm′s effectiveness is validated using both synthetic and real-world datasets, demonstrating its ability to uncover co-location patterns with high instance participation frequency, importance, and favorable distance distribution. However, it may miss a small number of co-location patterns with lower utility.

Key words: co-location pattern, high average-utility co-location pattern, feature average participation utility, pattern average
participation utility