A Unified Framework to Discover Partial Periodic-Frequent Patterns in Row and Columnar Temporal Databases

Partial periodic-frequent pattern mining is an important knowledge discovery technique in data mining. It involves identifying all frequent patterns that have exhibited partial periodic behavior in a temporal database. The following two limitations have hindered the successful industrial application of this technique: (i) there exists no algorithm to find the desired patterns in columnar temporal databases, and (ii) existing algorithms are computationally expensive both in terms of runtime and memory consumption. This paper tackles these two challenging problems by proposing a novel algorithm known as Generalized Partial Periodic-Frequent Depth-First Search (GPPF-DFS). The proposed algorithm compresses a given row or columnar temporal database into a unified dictionary structure and mines this structure recursively to find all desired patterns. Experimental results demonstrate that GPPF-DFS is 2 to 156 times faster and 5 to 88 times more memory efficient than the state-of-the-art algorithm. We also describe the usefulness of our algorithm with a case study on air pollution analytics.