Mining Significant Utility Discriminative Patterns in Quantitative Databases

Abstract

Drawing a discriminative pattern in quantitative datasets is often represented to return a high utility pattern (HUP). The traditional methods output patterns with a utility above a pre-given threshold. Nevertheless, the current user-centered algorithm requires outputting the results in a timely manner to strengthen the interaction between the mining system and users. Pattern sampling can return results with a probability guarantee in a short time, and it could be a candidate technology to mine such discriminative patterns. In this paper, a novel approach named HUPSampler is proposed to sample one potential HUP, which is extracted with probability significance according to its utility in the database. HUPSampler introduces an interval constraint on the length of HUP and randomly extracts an integer k according to the utility proportion firstly; then, the HUPs could be obtained efficiently from a random tree by using a pattern growth way, and finally, it returns a HUP of length k randomly. The experimental study shows that HUPSampler is efficient in regard to memory usage, runtime, and utility distribution. In addition, case studies show that HUPSampler can be significantly used in analyzing the COVID-19 epidemic by identifying critical locations.