基于蝙蝠算法的贝叶斯分类器优化研究  

ON BAYESIAN CLASSIFIER OPTIMISATION BASED ON BAT ALGORITHM

作　　者：蒋礼青 张明新[2] 郑金龙[2] 戴娇[1] 

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221116 [2]常熟理工学院计算机科学与工程学院,江苏常熟215500

出　　处：《计算机应用与软件》2016年第9期259-263,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61173130)

摘　　要：朴素贝叶斯分类器是一种应用广泛且简单有效的分类算法,但其条件独立性的"朴素贝叶斯假设"与现实存在差异,这种假设限制朴素贝叶斯分类器分类的准确率。为削弱这种假设,利用改进的蝙蝠算法优化朴素贝叶斯分类器。改进的蝙蝠算法引入禁忌搜索机制和随机扰动算子,避免其陷入局部最优解,加快收敛速度。改进的蝙蝠算法自动搜索每个属性的权值,通过给每个属性赋予不同的权值,在计算代价不大幅提高的情况下削弱了类独立性假设且增强了朴素贝叶斯分类器的准确率。实验结果表明,该算法与传统的朴素贝叶斯和文献[6]的新加权贝叶斯分类算法相比,其分类效果更加精准。Naive Bayesian classifier is a widely used simple and efficient classification algorithm, but there is a difference between the ＂naive Bayesian hypothesis＂ of conditional independence and the reality, such hypothesis restricts the accuracy of naive Bayesian classifier. To cripple this hypothesis, we used the improved bat algorithm （IBA） to optimise naive Bayesian classifier. IBA introduces Taboo search mechanism and random perturbation operator to prevent the naive Bayesian classifier from falling into local optimal solution, and thus socelerate, the convergence rate. IBA automatically searches the weight value of every attribute, by assigning different weight to each attribute, it weakens the hypothesis of independence and enhances the accuracy of naive Bayesian classifier without greatly increasing the cost of calculation. Test result showed that comparing with traditional naive Bayesian and newest weighted Bayesian classification algorithm proposed in paper[6], the proposed algorithm has higher accuracy in classification effect.

关 键 词：分类 朴素贝叶斯 属性加权 蝙蝠算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]