Boosting first-order clauses for large, skewed data sets

Authors:
Louis Oliphant;Elizabeth Burnside;Jude Shavlik
Affiliations:
Computer Sciences Department and Biostatistics and Medical Informatics Department, University of Wisconsin-Madison;Radiology Department and Biostatistics and Medical Informatics Department, University of Wisconsin-Madison;Computer Sciences Department and Biostatistics and Medical Informatics Department, University of Wisconsin-Madison
Venue:
ILP'09 Proceedings of the 19th international conference on Inductive logic programming
Year:
2009

Citing 8
Cited 0

Relational learning and boosting

Relational Data Mining
Ensemble Methods in Machine Learning

MCS '00 Proceedings of the First International Workshop on Multiple Classifier Systems
An Efficient Boosting Algorithm for Combining Preferences

ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
The relationship between Precision-Recall and ROC curves

ICML '06 Proceedings of the 23rd international conference on Machine learning
Gleaner: Creating ensembles of first-order clauses to improve recall-precision curves

Machine Learning
Representing sentence structure in hidden Markov models for information extraction

IJCAI'01 Proceedings of the 17th international joint conference on Artificial intelligence - Volume 2
An empirical evaluation of bagging in inductive logic programming

ILP'02 Proceedings of the 12th international conference on Inductive logic programming
An integrated approach to learning bayesian networks of rules

ECML'05 Proceedings of the 16th European conference on Machine Learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

Creating an effective ensemble of clauses for large, skewed data sets requires finding a diverse, high-scoring set of clauses and then combining them in such a way as to maximize predictive performance. We have adapted the RankBoost algorithm in order to maximize area under the recall-precision curve, a much better metric when working with highly skewed data sets than ROC curves. We have also explored a range of possibilities for the weak hypotheses used by our modified RankBoost algorithm beyond using individual clauses. We provide results on four large, skewed data sets showing that our modified RankBoost algorithm outperforms the original on area under the recall-precision curves.