Two-layered ensemble Kohonen nets for imbalanced streaming data

One of the dominant application areas of machine learning will be the industrial internet (Internet of Things), although most of the data will actually be confined to smaller internal networks and will not use the Internet. In this world of the industrial internet, streaming sensor data with rare events, such as machine or process failures and disruptions, will be the dominant form of data and it poses a particular problem for machine learning because its methods are generally not designed to handle imbalanced data. In this paper, I present a new classification algorithm for streaming imbalanced data that uses Kohonen nets at its core. To handle high-velocity streaming data, the proposed algorithm can be implemented on hardware to take advantage of massive parallelism. I provide an outline of the algorithm and some preliminary computational results.