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.

中文翻译：

---

基于粒计算的多路径选择分层分类

由于具有分层类结构的数据的广泛存在，分层分类是机器学习中的一个研究热点。现有的基于粒计算的分层分类方法可以通过考虑类的粒度来有效降低计算复杂度。然而，它们的预测准确性受到层次结构内层间错误传播的影响。在本文中，我们提出了一种基于粗粒度和细粒度类关系的多路径选择分层分类方法，该方法减轻了层间错误传播问题。首先，我们使用自顶向下的递归方法通过逻辑回归分类计算层次类的概率。第二，当前类概率是通过组合父类概率和当前类概率来计算的。我们根据它们的兄弟关系在当前级别选择多个可能的细粒度类。与现有方法相比，所提出的方法降低了上层误分类的可能性。最后，将多径预测结果提供给经典分类器进行最终预测。我们的分层分类方法在六个基准数据集上进行了评估，以证明它提供了比现有最先进的分层方法更好的分类性能。所提出的方法降低了上层错误分类的可能性。最后，将多径预测结果提供给经典分类器进行最终预测。我们的分层分类方法在六个基准数据集上进行了评估，以证明它提供了比现有最先进的分层方法更好的分类性能。所提出的方法降低了上层错误分类的可能性。最后，将多径预测结果提供给经典分类器进行最终预测。我们的分层分类方法在六个基准数据集上进行了评估，以证明它提供了比现有最先进的分层方法更好的分类性能。