Predicting recurring concepts on data-streams by means of a meta-model and a fuzzy similarity function

摘要: Implementation of a new similarity concept function using fuzzy logic techniques.Implementation meta-model representing the underlying drifts detected.Development MM-PRec as wrapper mechanism.Useful in intrusion detection systems.Useful fraud systems. Stream-mining approach is defined set cutting-edge techniques designed to process streams data real time, order extract knowledge. In particular case classification, stream-mining has adapt its behavior volatile distributions, what been called drift. It important note that drift may lead situations where predictive models become invalid and have therefore be updated represent actual concepts poses. this context, there specific type drift, known recurrent represented by already appeared past. those cases learning could saved or at least minimized applying previously trained model.To deal with aforementioned scenario, meta-models can used enhancing mechanisms stream algorithms, predicting when change will occur. There are some real-world reappears, systems (IDS), same incidents an adaptation them usually reappear over time. these environments early prediction means better knowledge past help anticipate change, thus improving efficiency model regarding training instances needed.Furthermore, complement meta-models, mechanism assess between classification also needed dealing concepts. reusing rough comparison made, boolean logic. The introduction comparisons efficient reuse seen concepts, not just equal models, but similar ones.This work faces open issues system, integrates function. theoretical proposal validated paper different experiments both synthetic datasets.