Implementation and Analysis Optimal Flexible Frequency Discretization (OFFD) Method to Minimize Classification Error at Naïve Bayes Classification

Abstract

Naive Bayes is one of the classification techniques in data mining that apply Bayes Theorem in its processing and provide optimal result when each attributes in dataset is independent. But generally, a dataset has numeric attributes and nominal attributes are dependence so that if considered independent, it can cause classification error problems. Therefore, it needs a method to minimize the error rate, the method is discretize strategy. Discretization is a method that maps some numerical values (X) into an interval of nominal value (X*) based on the frequency setting in one interval so it can get number of interval formed in one numeric attribute.

One of discretization method adopted in this research is Optimal Flexible Frequency Discretization (OFFD) based on sequential search and wrapper based supervised for incremental learning. This method will be carried out wrapper feature selection to get optimal attributes based on its fMeasure parameter. Then, optimal dataset will de discrete in sequential search for the minimum frequency on each interval. Based on the results of testing, showed that the OFFD influenced by the process of selecting attributes of Best First Search on the Wrapper Feature Selection, so that influence the decline in the value of the error.

Keywords : wrapper based, Feature Selection, discretization, sequential search, Naïve Bayes, Optimal Flexible Frequency Discretization,  interval frequency