Naïve Bayes Classification Using Kernel Independent Component Analysis

Nave Bayes is one of the most well-known efficient learning algorithms. It has been used extensively in a number of different areas of data mining and machine learning. It is known to have outstanding classification performance, which competitive with modern methods such as support vector machines (SVM). Nave Bayes classifier is used for building on independent features. Although the assumption of independence on which nave Bayes is based is rarely satisfied in practice, these features can be modified to make the independent assumption hold, which result in improving nave Bayes classifier. A number of current techniques have been used for that modification. Typically these methods find a set of uncorrelated/independent components of the features. Kernel independent component analysis (KICA) is a popular example of such methods. This paper proposes an application of kernel independent component analysis (KICA) with nave Bayes for classification. Kernel independent component analysis (KICA) is a kernel version of independent component analysis (ICA). Applying KICA yields new reduced independent features space on which the nave Bayes classifier can be built. For assessing the classification performance of this technique, this method is compared with independent component analysis (ICA) and kernel principle component analysis (KPCA) as a kernel approach. A number of performance measures have been used in the comparison; accuracy, sensitivity, specificity, and precision, F-score, the area under receiver operating characteristic curve (AUC) and the receiver operating characteristic (ROC) analysis.