In this paper, we study the properties of the KCCE algorithm (Kernel-based Classification with Cluster Ensemble) proposed in our previous publication. Different strategies for choosing the ensemble weights are investigated from the point of view of their influence on the quality and speed of the algorithm. We perform numerical experiments with the algorithm on real datasets from the UCI repository under conditions of artificially added noise at different levels.
|Journal||Journal of Physics: Conference Series|
|Publication status||Published - 4 Jan 2021|
|Event||International Conference on Marchuk Scientific Readings 2020, MSR 2020 - Akademgorodok, Novosibirsk, Russian Federation|
Duration: 19 Oct 2020 → 23 Oct 2020