| Summary: | 碩士 === 元智大學 === 資訊管理學系 === 98 === Decision tree inductive learning method is one of the popular classification techniques in machine learning. It uses training data to construct decision tree model, which is straightforward and interpretable. Thus, decision tree has been applied into many various domains. Decision tree is good at handling nominal data but not continuous data. It converts continuous-valued attribute to finite discrete numbers by determining the cutting points, and then picks one of the best cutting points to be the branch point. In this study, a novel algorithm incorporating PSO (Particle Swarm Optimization) and decision tree is applied to select the cutting point of continuous-valued attributes. It aims to generate more accurate classification rules faster. UCI-ML database and National Health Insurance Research database is the resource of testing data, and both the accuracy of prediction and the size of decision tree model will be evaluated.
In the result of experiment, accuracy of the decision tree construct from four UCI-ML dataset by our algorithm is better than other four popular decision tree algorithms, and the size of the decision tree construct from our algorithm has no apparent increase or less than other four popular decision tree algorithms. The decision tree construct from National Health Insurance dataset by our algorithm represent 21 classification rules of fracture, according to these rules we find 9 important factors of fracture.
|
|---|
