A Frequent Pattern Conjunction Heuristic for Rule Generation in Data Streams

This paper introduces a new and expressive algorithm for inducing descriptive rule-sets from streaming data in real-time order to describe frequent patterns explicitly encoded the stream. Data Stream Mining (DSM) is concerned with automatic analysis of streams real-time. Rapid flows challenge state-of-the art processing communication infrastructure, hence motivation research innovation into algorithms that analyse on-the-fly can automatically adapt concept drifts. To date, DSM techniques have largely focused on predictive mining applications aim forecast value particular target feature unseen instances, answering questions such as whether credit card transaction fraudulent or not. A real-time, technique has not been previously established part toolkit. motivated work reported this paper, which resulted developing validating Generalised Rule Induction (GRI) tool, thus producing rules explanations be easily understood by human analysts. The expressiveness decision models serves objectives transparency, underpinning vision ‘explainable AI’ yet an area attracted less attention despite being high practical importance. introduced described termed Fast (FGRI). FGRI able induce incrementally raw both categorical numerical features. changes pattern stream (concept drift) fly arrives applied continuously also provides theoretical, qualitative empirical evaluation FGRI.