Mining periodic patterns from floating and ambiguous time segments

A partial periodic pattern is referred to as a set of events that exhibits cyclic behavior over some periods in a time series. Previous studies focused on mining such patterns from constant-length segments that are always stationary between fixed offsets within periods. Unfortunately, many objects under study, like mice, seldom align their fluctuating behaviors with stationary segments that are created based on artificial concepts such as hours. As the result, many patterns that occur across artificial boundaries may not be detected. In this paper, we present a more flexible model that can dynamically create floating segments from a time series such that each segment captures one type of sporadic activity, such as running, over irregular length of time. Temporal sections that are frequently overlapped by floating segments are detected as 1-section patterns. Combinations of some 1-section patterns, referred to as K-section patterns, can then be discovered by the Apriori algorithm.