软聚类节点分裂层次模型

Aiming at the structural redundancy and error migration existing in the classification decision tree algorithm, a soft clustering node split hierarchical model was proposed. Through the decision-making model at the leaf nodes and the method of splitting nodes by soft clustering, the efficient partitioning of the sample space was realized, and a simplified hierarchical model was generated. Using the hierarchical discriminant method, samples were predicted with weighted summation methods from the leaf nodes to the root node of hierarchical structure model to reduce the effect of model structure on classification performance, and to improve the model's ability in discriminant errors adjustment. Compared with CART, ID3 and C4.5, the model proposed by the method is simple and showes the best classification performance on two data sets, F₁-measure is 0.53 and 0.38 respectively. The experimental result indicates the soft clustering node split hierarchical model can avoid the redundant structure and alleviate the problem of error migration.