Automatic learning of predictive rules for complex event processing

Abstract

International audienceThe inference mechanisms of CEP engines are completely guided by rules, which are specified manually by domain experts. We argue that this user-based rule specification is a limiting factor, as it requires the experts to have technical knowledge about the CEP language they want to use, it restricts the usage of CEP to merely the detection of straightforward situations, and it restrains its propagation to more advanced fields that require prediction and proactivity. Therefore, we introduce autoCEP as a data mining-based approach that automatically learns predictive CEP rules from historical traces. More precisely, we include our novel method that is capable of learning rules and handling events coming from one source, and then we elaborate our vision on how to extend autoCEP to deal with simultaneous events coming from multiple sources

Rule‐based preprocessing for data stream mining using complex event processing

“…A first proposal, iCEP, breaks down the problem of rule definition into several steps, such as identifying relevant event attributes or determining a proper window size (Margara et al, 2014). Similarly, autoCEP learns sequences from time series and transforms them into ready-to-deploy CEP rules (Mousheimish et al, 2017). Recently, GP4CEP proposes the use of genetic programming to build the rules that best describe the occurrence of a predefined complex event (Bruns et al, 2019).…”

A situation refinement model for complex event processing

“…These rules must be modified by human experts to identify and express correlations between the various events and situations of interest. Providing all the required information to perform rule updates manually is a hard task, especially with the need to perform it continuously with low latency [11,21].…”

“…Several research efforts have pointed out the challenge of applying data mining techniques to learn CEP detection rules in real-time streaming environments especially with limited time available to infer such rules from the event stream [18,19]. Existing rule learning approaches either focus on mining CEP rules from historical instances of events [20][21][22][23] or external domain knowledge [13,14]. Other attempts have been proposed to incorporate domain knowledge with historical events in the rule learning process such as the SCEP ter framework [24].…”

“…Despite partial success in automating CEP rule generation process either by finetuning existing CEP rules [23,25], or by generating entirely new CEP rules from historical events [20,21], dealing with the increased number of rules due to the unbounded nature of event stream is still a challenging task. Specifically, an increased number of CEP detection rules lead to more complex detection patterns, which makes the targeted situation more complicated to comprehend and handle [12,20,26].…”

“…Other attempts focused on combining domain knowledge with historical events in the rule learning process, such as the SCEP ter framework proposed by Zhou et al [24]. Despite the partial success in automating the CEP rule generation process either by fine tuning existing CEP rules [23,25] or by generating entirely new pattern rules [20,21], dealing with the increased number of rules is still a challenging task. Precisely, the increased number of CEP detection rules leads to more complex detection patterns, which makes the targeted situation more complicated to comprehend and handle [12,20,26]; furthermore, such an increase also brings higher levels of computational complexity [27].…”