Integrating advanced principal component analysis into naive bayes for enhanced classification performance

The Naive Bayes algorithm is one of the most important and popular algorithms in machine learning and data mining, not only because of its simplicity but also because of its superior classification performance. The central assumption of this algorithm is known as the attribute independence assumption. This assumption allows the Naive Bayes algorithm to solve classification problems conveniently, but also limits the performance of this algorithm to a certain extent when the mixed type of variables exist in its input dataset. Recently, we proposed an improved Naive Bayes classification algorithm by combining an improved Principal Component Analysis (PCA) method. The improved PCA first calculates correlation coefficients between coupling variables using the Pearson and Kendall coefficients, where the two types of coefficients are calculated separately for quantitative and qualitative data. After coupling data is transformed into principal components, those correlated variables can be integrated into the improved Naive Bayes algorithm. When the improved Naive Bayes algorithm is applied to a classified task, it is easy to verify that the transformed principal components data are approximately independent, thereby conforming to the Naive Bayes independence assumption to a relatively greater extent. This implies that it is likely for the improved Naive Bayes algorithm to yield a more accurate classification performance, as it is more robust to the presence of noise in classification instances.