A novel classification tree based on local minimum Gini index and attribute partial order structure diagram

Abstract

Decision tree is not only an important machine learning method, but also the basis of ensemble learning methods such as random forest and deep forest. Based on the theory of Formal Concept Analysis (FCA) and Attribute Partial Order Structure Diagram (APOSD), a new decision tree for classification is proposed in this paper. Firstly, the local minimum of Gini index is used to complete the data granulation, and the Formal Decision Mode Information Table (FDMIT) is constructed. Then, the Attribute Partial Order Classification Tree (APOCT) is generated based on APOSD to complete the pattern recognition and rule extraction. The method of APOCT separates the process of granulation and visualisation, and the granulation process is easy to parallelise and efficient. The experimental results show that APOCT is effective.