Hierarchical classification with multi-path selection based on granular computing

Hierarchical classification is a research hotspot in machine learning due to the widespread existence of data with hierarchical class structures. Existing hierarchical classification methods based on granular computing can effectively reduce the computational complexity by considering the granularity of classes. However, their predictive accuracy is affected by inter-level error propagation within the hierarchy. In this paper, we propose a hierarchical classification method with multi-path selection based on coarse- and fine-grained class relationships, which mitigates the inter-level error propagation problem. Firstly, we use a top-down recursive method to calculate the probabilities of the hierarchical classes by logistic regression classification. Secondly, the current class probability is calculated by combining the parent and current classes probabilities. We select multiple possible fine-grained classes at the current level according to their sibling relationships. Compared with existing methods, the proposed method reduces the possibility of misclassification from the upper layer. Finally, the multi-path prediction result is provided to a classical classifier for final prediction. Our hierarchical classification method is evaluated on six benchmark datasets to demonstrate that it provides better classification performance than existing state-of-the-art hierarchical methods.