A0363
Title: Classification trees with Gini samples splits
Authors: Amirah Alharthi - University of Leeds (United Kingdom) [presenting]
Abstract: Many numerical studies indicate that bagged decision stumps preform more accurately than a single stump. We will introduce a new stump-based ensemble method for classification which is: A forest of stumps `Gini-Sampled Splits'. A stump within this forest uses a split that is generated from transformed Gini indices for each possible cut points. The choice of variable which is chosen on which to generate a split has a probability proportional to that variable Gini index values. The final decision of these stumps is aggregated using weighted vote rather than majority vote. We compared between this method and other tree-based ensemble classification methods in terms of the accuracy and the results are promising.
