Classification data mining with Laplacian Smoothing on Naïve Bayes method

Abstract

Data mining is a process of gathering information to find important pattern recognition in the data set in the database so that it becomes knowledge discovery. Classification is a technique of grouping data based on data attachment to sample data. Naïve Bayes is one of the techniques in data mining classification that uses the probability method and is better known as the Naïve Bayes Classifier (NBC). The main characteristic of NBC is that there is a strong assumption of independence from each condition (independent variable). By applying Naive Bayes to a data sometimes causes misclassification if the training data is only a few so that the testing data is not found in the training data and this causes the probability result to be zero and an error in the classification process. To avoid zero probability results that cause errors in the classification process, a refinement method is needed. Laplacian Smoothing is a smoothing technique that can be used in Naïve Bayes classification in an easy way. The concept is to add a small positive value to each of the existing conditional probability values to avoid zero values in the probability model. Laplacian Smoothing on NBC is examined in this article.