ENSEMBLE ALGORITHM FOR DATA STREAMS WITH CONCEPT DRIFT

The ensemble algorithms have been particularly efficient to work on areas of large and complex data as so-called data streams. In these flows, during qualifying, they appear concepts change over time, so that its mining methods, especially those that detect and adapt to these changes are important for their application in areas such as bioinformatics, medicine, economics and finance, industry, environment, among others. This research proposes a new ensemble algorithm that adapts to concepts drift, have weighted voting with a new way to adjust the weights and can vary the type of basic classifier. The algorithm was implemented in support and under the demands of the work environment MOA (Massive Online Analysis) facilitating comparison with other known algorithms and the generation of synthetic data bases that simulate changes in concepts. For experimentation, experiences they were generated under known, such as artificial concepts: SEA, LED, STAGGER and hyperplane; managing to show high resilience and stability of the algorithm against different simulated situations.