Implementation of Data Stream Classification Neural Network Models Over Big Data Platforms

Streaming is being increasingly demanded because it helps in analyzing data in real-time and in decision making. Over time, the number of existing devices increases continuously, generating a huge amount of data. Processing this data with traditional algorithms is impractical, so it is necessary to apply distributed algorithms in a Big Data context. In this paper, Apache Spark is used to implement some distributed versions of algorithms based on Extreme Learning Machine (ELM). In addition, these algorithms are evaluated with different real and synthetic datasets by performing two experiments. The first one tries to demonstrate that the performance of the distributed algorithms is the same as that of the sequential versions. The second experiment is a study about the behaviour of the algorithms in the presence of concept drift, an important research area within streaming.