Very Fast C4.5 Decision Tree Algorithm
This paper presents a novel algorithm so-called VFC4.5 for building decision trees. It proposes an adaptation of the way C4.5 finds the threshold of a continuous attribute. Instead of finding the threshold that maximizes gain ratio, the paper proposes to simply reduce the number of candidate cut poi...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2018-04-01
|
| Series: | Applied Artificial Intelligence |
| Online Access: | http://dx.doi.org/10.1080/08839514.2018.1447479 |
| Summary: | This paper presents a novel algorithm so-called VFC4.5 for building decision trees. It proposes an adaptation of the way C4.5 finds the threshold of a continuous attribute. Instead of finding the threshold that maximizes gain ratio, the paper proposes to simply reduce the number of candidate cut points by using arithmetic mean and median to improve a reported weakness of the C4.5 algorithm when it deals with continuous attributes. This paper will focus primarily on the theoretical aspects of the VFC4.5 algorithm. An empirical trials, using 49 datasets, show that, in most times, the VFC4.5 algorithm leads to smaller decision trees with better accuracy compared to the C4.5 algorithm. VFC4.5 gives excellent accuracy results as C4.5 and it is much faster than the VFDT algorithm. |
|---|---|
| ISSN: | 0883-9514 1087-6545 |