Weighted proportional k-interval discretization for naive-Bayes classifiers

摘要: The use of different discretization techniques can be expected to affect the classification bias and variance naive-Bayes classifiers. We call such an effect variance. Proportional k-interval (PKID) tunes by adjusting discretized interval size number proportional training instances. Theoretical analysis suggests that this is desirable for However PKID sub-optimal when learning from data small size. argue because equally weighs reduction reduction. But data, contribute more lower error thus should given greater weight than Accordingly we propose weighted (WPKID), which establishes a suitable trade-off while allowing additional used reduce both Our experiments demonstrate classifiers, WPKID improves upon smaller datasets with significant frequency; delivers significantly often not in comparison three other leading alternative studied.