Tensor decision trees for continual learning from drifting data streams

Data stream classification is one of the most vital areas contemporary machine learning, as many real-life problems generate data continuously and in large volumes. However, research this area focuses on vector-based representations, which are unsuitable for capturing properties more complex multi-dimensional structures, such images video sequences. In paper, we propose a novel methodology learning adaptive decision trees from streams tensors. We introduce Chordal Kernel Decision Tree continual tensor streams. order to maintain characteristics, train update classifiers kernel space designed work with representation. use chordal distance compute similarities between tensors then apply it new feature trained. This allows direct tree induction accommodate streaming drifting nature data, concept drift detection scheme based It us reconstruct every time when change detected. The proposed approach fast efficient Experimental study, conducted 4 real-world 52 artificial large-scale streams, shows that using native leads accurate than outperforms vectorized representations.