利用基因規劃法改進建立決策樹的方法

Abstract

決策樹已在相當多的領域廣泛被使用，因為它的易於表達且易於建構。雖然基因規劃法的發展歷史並不長，但已有許多論文在討論它的應用。本篇論文便提出一個模型，利用基因規劃法來產生決策樹，並利用實作來分析該決策樹應用層面及效率。 本論文提出的模型中，利用了基因規劃法以演化的方式建立決策樹。並力求模型的簡單易瞭。同時我們也針對因訓練資料具有高分支度的屬性而影響決策樹的建立的情況提出改進的方法。 於本篇論文中，我們利用打網球的例子來實作。為了更加驗證本論文所提的方法，我們也實際使用銀行信用卡授信額度作測試，以強調此模型的實用性。

Decision trees have been performed well in many fields because they are easily to be expressed and inducted. In despite of short developing time, quite a few papers have discussed the applications of decision trees. In this thesis, we propose a model in using the genetic programming paradigm to induct decision trees, and analyze these trees' performance by applying implementation. In this thesis, we use the Genetic Programming paradigm to induct decision trees. We try to simplify the model to be easily understood. Also we offer a method to reduce the influence of training set's highly branched attribute. In this thesis, we prove the usefulness of this proposed model by applying a tennis game. For verifying the proposed model in this thesis, we offered decisions of credit limit of the credit cards to testify to the application of this proposed model.