Abstract：Naive Bayes(NB) classifier has exhibited excellent performance on many problem domains due to its simplicity and efficiency. In reality the conditional independence assumption of Nave Bayes isn’t always true. Attribute weighting is one of the most popular methods to alleviate this assumption’s influence on classification results. However, traditional classification models ignore characteristics of each test instance, and the weight vector learned from the whole training set failed to reflect each attribute’s contribution of distinguishing each test instance correctly. To this end, a data driven lazy learning locally attribute weighted naive Bayes model is proposed. The attribute weights for each test instance are learned from its neighborhoods, and learned weights are employed to build the locally weighted model by optimization method. Experimental results on benchmark datasets demonstrate that the proposed approach is more accurate than other classical classifiers.