부산대학교 도서관

Summary: Classification trees are an extensively researched solution to classification problems. Pruning is a very important element in constructing a tree, since it determines the final tree. Generally, a shorter, simpler tree is preferred if it maintains the accuracy of the tree. In this dissertation, a new pruning algorithm is proposed. Comparison with other existing methods is performed on both simulated data and real data. Results show that our proposed method provides high accuracy, fast execution times and comparable tree sizes. An imputation method is proposed to deal with missing data. Results show that our method works well in the presence of missing values.