Introducing Confidence as a Weight in Random Forest
Authors:
- 1 January 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 611-616
Abstract
Random forest is one of the most popular supervised learning methods which is a collection of multiple decision trees. It is computationally fast, easy to use and gives reasonable performances over diversified applications. In a random forest, outcomes from multiple trees are aggregated to make the final decision where all trees get the same importance. However, it is natural that the performance of all trees is not equal and thus calculating the majority vote from these trees may not always give the correct decision. To improve this situation, we propose Confidence weighted Random Forest (CwRF) which uses a confidence factor for each leaf node in every decision tree and use that confidence to take weighted majority vote for making the final decision. A rigorous experiment over twenty-five datasets illustrates that the proposed CwRF outperforms three other state-of-the-art methods and provides significantly better performance over nineteen datasets out of twenty-five datasets at best.Keywords
This publication has 16 references indexed in Scilit:
- Face Alignment at 3000 FPS via Regressing Local Binary Features2014
- A weighted random forests approach to improve predictive performanceStatistical Analysis and Data Mining: An ASA Data Science Journal, 2013
- Class-Specific Hough Forests for Object Detection2013
- Random feature weights for decision tree ensemble constructionInformation Fusion, 2012
- A weight-adjusted voting algorithm for ensembles of classifiersJournal of the Korean Statistical Society, 2011
- Mining data with random forests: A survey and results of new testsPattern Recognition, 2011
- On Oblique Random Forests2011
- Extremely randomized treesMachine Learning, 2006
- Ensembling neural networks: Many could be better than allArtificial Intelligence, 2002
- A survey of decision tree classifier methodologyIEEE Transactions on Systems, Man, and Cybernetics, 1991