An Improved Bayesian Learner Based on Weighted Beta Kernel Density Estimation

There are problems in classification tasks where noisy data situations can affect the probability density fitting, as well as adaptive problems with current Bayesian classifiers constructed with symmetric kernel density estimation. In this paper, we will propose a plain Bayesian classifier improved based on weighted beta kernel density estimation. The bandwidth selection is also derived for beta kernel estimation. Compared with the classifier based on symmetric kernel density estimator, the beta kernel can be more adaptive to data changes and can solve certain boundary influence problems. Meanwhile, for the presence of noisy data in the data, a weighted kernel density estimation is combined with the beta kernel to weaken the influence of some of noisy data in fitting the data distribution. The experimental results show that our proposed weighted beta kernel density estimation plain Bayesian classifier method has obvious effect and improvement in fitting probability density distribution and data classification for tightly supported data.