- About
- Step 1 - User Prompted to Select Input File
- Step 2 - Decision Tree Displayed
- Step 3 - Query Tuple Provided
- Step 4 - Prediction
- Demo
Decision Trees are a way of finding correlations in people's behavior from data about their habits. This program builds a decision tree based on a given input file and then proceeds to classify a user-provided tuple of data. In the example below, the input file holds information that a computer manufacturing company has collected about their customers. The Decision Tree Generator will attempt to find other factors and combinations of factors (such as age, income, and student status) related to that a customer's is likeliness to make a purchase.
The program prompts the user for choose an input file containing the data that is to be used for generating a decision tree.
The program uses Java's Jfilechooser object for allowing the user to choose the input file through a GUI.
This is sample information that a computer manufacturing company has collected about their customers. The first row of the input file provides the names of attributes while the rest of the rows are collected data points corresponding to those attributes. In this example input file the attributes are:
- age
- income
- student
- credit_rating
- buys_computer
Reading the second row, we can see it belongs to a customer who falls in the youth age class, the high income class, the no student class, the fair credit rating class, and the no buys_computer class. So this person is young, makes high income, is not a student, has a fair credit rating, and did not buy a computer.
Here, the decision tree is displayed in a JTextArea object. The root of the tree will be the attribute with the highest calculated entropy and is used to produce the first split in the tree. Entropy is an estimate of the information that could be gained by splitting that decision tree using a given attribute. In the image below, we can see that the attribute with the highest calculated entropy was age and the possible classes include youth, middle_aged, or senior. The attribute with the highest entropy is selected in order to keep the tree as small as possible.
User provides a data point in the form of a tuple, excluding the value for the attribute that is being predicted. The tuple provided here is young, makes high income, is a student, and has fair credit.
A simple prediction is displayed in the message dialog box which indicates the class prediction for the attribute buys_computer.
