Multi-Objective sequential pattern mining over transaction data streams based on a sliding window model

Abstract

Sequential pattern mining plays an essential role in numerous practical domains such as customer behavior analysis, medical diagnosis, and cybersecurity monitoring. Although previous studies have effectively addressed the problem of discovering frequent patterns or high utility patterns in static databases, the rapid growth of continuously updated data streams has raised critical challenges regarding processing latency and storage limitations. To overcome these obstacles, this paper proposes the Multiple Objective Stream Sequential Pattern Mining algorithm (MOS-SPM) that operates on a specialized linked list structure (MOS-SPM-LL) based on a sliding window model. This approach enables the system to simultaneously evaluate three core criteria, including support, high utility, and regularity of patterns, directly over a transaction stream. For comprehensive evaluation, the study proactively extends two state-of-the-art algorithms, CHUSP and RStreamHUSP, as comparison baselines. Experimental results demonstrate that the proposed algorithm preserves absolute correctness when compared with the extended CHUSP version. Moreover, owing to its lightweight storage mechanism, MOS-SPM exhibits superior performance in both execution speed and memory optimization compared with the extended RStreamHUSP version.