L^2DLNB：懒惰学习双层朴素贝叶斯分类器

尽管朴素贝叶斯简单而且在很多数据集上效果很好，但是其属性独立性假设在现实世界中并不总是成立的，当这一假设不成立时，其结果很差。通过分析和研究，提出了一种放宽这种独立性假设的新算法——懒惰学习双层朴素贝叶斯分类器L^2DLNB，该算法使用基于条件互信息的懒惰学习方法，在求不同类标的似然度时，使用不同的属性依赖关系，从而能够更准确地计算出各类标似然度。实验结果表明此算法在一些数据集上取得了更好的分类精度。

Lazy Learning Based Double Layer Naive Bayesian Classifier

Though nave Bayesian classifier is simple and has good performance on many data sets, its attribute independence assumption does not always exist in the real world. Its performance is poor while the assumption is violated. In order to relax this assumption, L2DLNB( Lazy Learning Based Double Layer Nave Bayesian classifier ), is proposed, which could accurately calculate the likelihood, using condition mutual information based lazy learning, and different attribute dependent relation when to calculate the likelihood of different label. Experimt results indicate that L2DLNB improves classifier accuracy on some datasets compared to other Bayesian classifiers.