Temporal importance and utility based pattern analysis with negative units for industrial data streams

Assessing profitable aspects is an essential component across industrial domains. Prior research analyzing high utility patterns from data streams extracts profitable patterns with negative profits. However, it analyzes such streams without considering the temporal importance, posing challenges in reflecting the current trends. To overcome these limitations, a novel temporally aware approach is proposed for data streams. This method utilizes a time-decay approach to analyze patterns with positive and negative utility values. The proposed approach places higher importance on newly arriving transactions to analyze various profitable aspects for industrial domains. Additionally, novel pruning strategies based on the time-decay mechanism reduce the search space in the expansion process, improving efficiency without any loss of results. Extensive experiments covering runtime, memory usage, and scalability show that the proposed approach achieves up to 4 times faster runtime and 1.5 times better memory usage compared to state-of-the-art approaches. Moreover, statistical evaluations indicate the proposed method extracts fewer and more refined results without any loss, and a pattern quality evaluation demonstrates that it analyzes results reflecting temporal trends.