Decision Tree Induction Based on Reinforcement Learning Modelling and its Application on State Space Partition

• Main Authors: Tsung-Wen Yang, 楊琮文
• Other Authors: Kao-Shing Hwang
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/60487693075925935084

碩士 === 國立中正大學 === 電機工程研究所 === 93 === Most of the tree induction algorithms are typically based on a top-down greedy strategy that makes locally optimal decision at each node. This strategy may induce a larger tree that requires more tests. In this thesis, a decision tree induction which is based on reinforcement learning is proposed to avoid the greedy problem. The basic idea is that the splitting criterion is based on the long-term evaluations of splits instead of the local evaluations. The induction problem is modeled as reinforcement learning problem and solved by the technique of the domain. The method consists of two phases: split estimation and tree growing. In split estimation phase, the inducer estimates the long-term evaluations of splits at visited nodes. In the second phase, the inducer grows the tree using the learned long-term evaluations. A comparison with CART on several datasets is reported. The proposed method is then applied to tree-based reinforcement learning. The decoder in AHC is replaced by a regression tree, which is constructed by the proposed method. The experimental results are also demonstrated