We define a formula for estimating the coding costs of decision lists for propositional domains. This formula allows for multiple classes and both categorical and numerical attributes. For artificial domains the formula performs quite satisfactory, whereas results are rather mixed and inconclusive for natural domains. Further experiments lead to a principled simplification of the original formula which is robust in both artificial and natural domains. Simple hill-climbing search for the most compressive decision list significantly reduces the complexity of a given decision list while not impeding and sometimes even improving its predictive accuracy.
CITATION STYLE
Pfahringer, B. (1997). Compression-based pruning of decision lists. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 1224, pp. 199–212). Springer Verlag. https://doi.org/10.1007/3-540-62858-4_85
Mendeley helps you to discover research relevant for your work.