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

Data preprocessing is known to be essential produce accurate data from which mining methods are able extract valuable knowledge. When constantly arrives one or more sources, techniques need adapted efficiently handle these streams. To help domain experts define and execute tasks for streams, this paper proposes the use of active rule-based systems and, specifically, complex event processing (CEP) languages engines. The main contribution our approach formulation procedures as detection rules, expressed in an SQL-like language, that provide a simple way manipulate temporal data. This idea materialized into publicly available solution integrates CEP engine with library online mining. evaluate approach, we present three practical scenarios rules preprocess streams aim adding information, transforming features handling missing values. Experiments show how effective language express modular high-level manner, without significant time memory overheads. resulting do not only improving predictive accuracy classification algorithms, but also allow reducing complexity decision models needed learning some cases.