Kernel density estimation with stream data based on self-organizing map

We investigate the kernel density estimation (KDE) problem with stream data in this paper. Specifically, we analyze the characteristics of stream data density estimation, and propose an approach based on self-organizing map (SOM) to tackle the challenges of traditional KDE techniques for stream data analysis, such as computational cost, processing time, and memory requirement. Our proposed approach first generates SOMs for chunks of the data along the data streams, which obtains summaries of the data streams. Then, the probability density functions (pdfs) over arbitrary time periods along the data streams can be estimated with the generated SOMs. We compare our method with two other data stream KDE methods, the M-kernel and cluster kernel methods, in terms of accuracy and processing time. The simulation results illustrate the effectiveness and efficiency of the proposed algorithm.