II. Literature Review
2.2. Research Definition and Framework of Multi-staged Binary Tree
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
classification method when the training dataset is small.
Another disadvantage of Decision Tree is that the validity period of Decision Tree is not explicitly specified as time dependent data may have some invalid rules.
It has to separate the rules into dependent and independent times attribute (Chattamvelli, 2009).
2.2. Research Definition and Framework of Multi-staged Binary Tree
2.2.1. Concept
Multi-staged Binary Tree is an extension method of the Decision Tree. Their concepts are similar but a difference between them is the procedure of classification. The number of the Multi-staged Binary Tree is based on the number of classes. If the number of the target classes was n, then the number of the Multi-staged Binary Tree would be n-1.
2.2.2. Methodology
In Multi-staged Binary Tree, each node splits into at most two child nodes (Chattamvelli, 2009). It classifies one dataset first and then classifies another dataset afterwards, which is the two-step classification while Decision Tree classifies all the datasets into the classes once. Each step of the process classifies for two classes: terminal node and non-terminal node. For an example of Figure 3, it classifies Class X into Class A and Non-Class A at the first step, and then classifies the non-terminal node of Non-Class A into Class B and Class C at the second step.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 3: The Representation of Multi-staged Binary Tree
Figure 4: Example of the Multi-staged Binary Tree Model3 (Resource from Fu & Tu, 2011)
3 Example of the Multi-staged Binary Tree Model is referenced from Intelligence on Gift Giving Website Based on Hybrid Approach of Decision Tree and Similarity by Fu & Tu (2011)
Root
Class A:4 Non-Class A:6
Node 1
Class A:4 Non-Class A:0
Node 2
Class B:3 Class C:3 Non-Class BC:0
Node 3
Class B:3 Class C:0 Non-Class BC:0
Node 4
Class B:0 Class C:3 Non-Class BC:0
Class X
Class A Non-Class A
Class B Class C
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
For the instance of Multi-staged Binary Tree classification in Figure 4, if there are 3 classes: A, B and C to be classified, it will be classified into Class A and Non-Class A at the first step and then classifies Non-Class A into Class B and Class C at the second step as Non-Class A includes Class B and Class C. In the results from an example of the model shows that the Class A is node 1, node 3 as Class B, and the node 4 is classified as Class C.
As this research is based on dynamic survey, using Multi-staged Binary Tree has greater advantage than Decision Tree when user searches for the desired place or plan on tourism recommendation system. When using Multi-staged Binary Tree model, users do not have to complete the whole questionnaire in order to be able to search, but they do under Decision Tree. It will be more efficient if users only need to fill in parts of questionnaire of the often extensive tourism survey.
In this study, the researcher used three approaches to generate the target trees by Multi-staged Binary Tree method, and the classification procedure refers to the procedure in Figure 4: (1) for “excited with unexcited” target category approach, the first step is to classify excited and unexcited dataset into the nodes of excited and unexcited, which unexcited is a non-terminal node. The second step is to continue classifying the non-terminal node of unexcited into the nodes of usable and warm. (2) For “usable and non-usable” target approach, the first step is to classify the dataset of usable and non-usable into the nodes of usable and non-usable, which non-usable is a non-terminal node. The second step is to continue classifying the non-terminal node of non-usable into the nodes of excited and warm.(3) for “warm and non-warm” target approach, the first step is to classify warm and non-warm dataset into the nodes of warm and non-warm, which
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
non-warm is a non-terminal node. The second step is to classify the non-terminal node of non-warm into excited and usable perceptions.
2.2.3. Strength and Weakness
Multi-staged Binary Tree and original Decision Tree are highly similar;
hence Multi-staged Binary Tree has the same advantages and weaknesses as Decision Tree. However, as Multi-staged Binary Tree is an improved version that was extended from Decision Tree, it is more accurate. People use Multi-staged Binary Tree more when applying to areas of complex schemes as the accuracy rate is more accurate and it can break up complex decisions into simpler decisions for the obtaining of more desirable solutions. Therefore, it shows that the more flexible the method, the greater the efficiency.
However, some may argue that Multi-staged Binary Tree is less favourable than Decision Tree due to the complexity of the method itself. Due to the number of large classes generated, it may have repeated nodes produced in the tree which causes the number of terminal nodes to be larger than the number of actual classes.
Thus, it would increase the search time and the memory space in the system (Safavian & Landgrebe, 1991). Overall, the performance of both Multi-staged Binary Tree and Decision Tree are dependent on the designing of the tree.