Improvement and Research of Naive Bayes Classification Based on Unbalanced Data Sets

When the sample distribution of each class in the training set is uneven and the data is sparse, the classification of naive Bayes algorithm is not accurate enough [1]. To solve this problem, an improved method based on data level and algorithm level is proposed in this paper to solve the problem that naive Bayes classifier tends to favor large classes while ignoring small classes when the sample distribution of each class in the training set is uneven. In order to better deal with the classification of unbalanced data sets, the oversampling algorithm is improved at the data level. The risk function of the naive Bayes classifier is calculated by the cost function on the improved data set. By constructing cost function adapted to data distribution and combining with naive Bayes algorithm based on cost sensitivity, binary unbalanced data is classified. The experimental results show that the improved algorithm is better than the traditional naive Bayes classification algorithm when the distribution of sample sets is not balanced.