基于不同概率密度估计方法的朴素贝叶斯分类器

贝叶斯方法是分类技术中的一个较为基本的方法,通过贝叶斯法则以及相关性质推导出贝叶斯分类器,而引入属性条件独立性假设后,我们得到了朴素贝叶斯分类器,在不降低分类器性能的基础上降低了计算复杂度,在实际应用中更为简便。不同的概率密度估计方法对于每个属性条件概率的估计效果不同,对朴素贝叶斯分类器的性能也有所影响。在本文中我们提出了两种概率密度估计方法：核密度估计和混合高斯。这两种方法各有优势和不足,我们将其应用在实例中,选取苏东41-33区块下古气井的89口测井曲线作为研究数据,分别用核密度估计和混合高斯对训练数据进行概率密度估计,并用单高斯模型作为对照,然后用朴素贝叶斯方法对测试数据进行岩性分类,并统计不同概率密度估计方法下的分类其性能即准确率。

Naive Bayes classifier based on different probability density estimation methods

Bayesian method is an essential method in classification technology, Bayesian classifier is deduced by Bayesian rule and related properties. After introducing the attribute conditional independence assumption, we obtain the Naive Bayesian classifier, which reduces the computational complexity without reducing the performance of the classifier, and is simpler in practical application. Different probability density estimation methods have different effects on the conditional probability of each attribute, and also affect the performance of Naive Bayesian classifier. In this paper, we propose two kinds of probability density estimation methods: kernel density estimation and mixed Gaussian model. The two methods have their advantages and disadvantages, and we apply it to the example. We select the 89 logging logs from the ancient gas wells in the 41-33 block of Su Dong as the research data. We use the kernel density estimation and the mixed Gaussian model to get the probability density estimation of the training data respectively, and the single Gaussian model is used as the contrast method. Then the Naive Bayesian method is used to classify the test data and the accuracy of the classification under different probability density estimation methods is calculated.