Machine Learning (CS 536)
Project Report
Wei Wang
ww@paul.rutgers.edu
December 15, 1997
Abstract:
We introduce several weak learning algorithms designed
for Ensembles of classifiers which includes Boosting and Bagging algorithm.
Comparative results on a substantial number of datasets are reported. It
turns out that Boosting surpasses Bagging algorithm over most of the datasets
in experiments. Some possible variations of Boosting algorithm are also
introduced in this report.
-
Introduction
-
The Learning Algorithms
-
The Bagging Algorithm
-
The Boosting Algorithm
-
The Weak Learning Algorithms
-
Algorithm FindAttrTest
-
SigmoidPerceptron and Randomized K-SigmoidPerceptron
-
Algorithm PairAttrTest
-
The Hybrid Boosting
-
The Experiments
-
Performence of Weak Learning Algorithms
-
Comparation Between SLP Itself and SLP Boosting
-
Comparation Between Boosting and Bagging
-
Comparation Among Boosting of Various Weak Learning Algorithms
-
Comparation Between Boosting and Hybrid Boosting
-
Future Work
-
References
-
Appendix: Source Codes (3,500 lines)
-
Makefile
-
bagging.h
-
bagging.c
-
boost.h
-
boost.c
-
hyboost.h
-
hyboost.c
-
hyb_bst.c
-
dataset.h
-
dataset.c
-
resamp.h
-
resamp.c
-
svt.h
-
svt.c
-
svt_pre.c
-
svt_bag.c
-
svt_bst.c
-
pat.h
-
pat.c
-
pat_bag.c
-
pat_bst.c
-
slp.h
-
slp.c
-
slp_bag.c
-
slp_bst.c
-
rkp.h
-
rkp.c
-
rkp_bag.c
-
rkp_bst.c
-
Contents
-
List of Tables
-
List of Figures
Return to Projects' Web Page