Feature engineering for a gene regulation prediction task

ABSTRACT

This paper describes an approach that won honorable mention for the gene regulation prediction task of the 2002 KDD Cup competition [1]. Our methodology used extensive cross-validation to direct the search for an appropriate problem representation and the selection of an 'off-the-shelf' induction algorithm. A prominent trait of the dataset is the presence of three hierarchical attributes, for each of which we generated a novel predictive feature: the percentage of positives hierarchically aggregated at the node specified by the instance.

REFERENCES

• 1: Mark Craven, The genomics of a signaling pathway: a KDD Cup challenge task, ACM SIGKDD Explorations Newsletter, v.4 n.2, p.97-98, December 2002 [doi>10.1145/772862.772877]
• 2: George Forman, An extensive empirical study of feature selection metrics for text classification, The Journal of Machine Learning Research, 3, 3/1/2003
• 3: Forman, G. A Method for Discovering the Insignificance of One's Best Classifier and the Unlearnability of a Classification Task. DMLL Workshop, ICML, 2002.
• 4: Vladimir N. Vapnik, The nature of statistical learning theory, Springer-Verlag New York, Inc., New York, NY, 1995
• 5: Weka machine learning project, www.cs.waikato.ac.nz/ml
• 6: Zhang, B. Is the Maximal Margin Hyperplane Special in a Feature Space? Hewlett-Packard Labs Tech Report HPL-2001-89, 2001.

INDEX TERMS

Keywords:
bioinformatics, hierarchical attributes, machine learning