Training Data Selection for Improving Performance of Voting Feature Interval 5 (VFI 5).

Abstract

Voting Feature Interval 5 (VFI 5) is a supervised algorithm and an inductive learning algorithm for inducing knowledge classification from training information. VFI 5 algorithm is capable of classifying sample very well and can provide an explanation why and how the class groups of new samples from the classification can be predicted in the individual vote that each feature has been assigned to the class. VFI 5 algorithm determines the point interval for the classification process. Point interval is obtained by taking the lowest and the highest value of the sample in each class. In the testing process, if the test data are outside the sample interval they will have zero voting value and will reduce the accuracy of the classification results. The selection of training data with non-random sampling method uses the purposive sampling technique. The selection process is done by taking a few of the lowest and the highest feature values from each feature data to be training data. The remaining data which are not used as training data will be used as testing data. The propotion of training and testing data is 2:1. Among the three data used in the VFI 5 algorithm with the selection training data using the lowest and highest feature values, the iris data produced an accuracy of 98.04%, the accuracy of wines data is 96.56% and the acuracy of gender koi data is very high reaching 100%. The result of this study shows that the algorithm VFI 5 data selection method using the lowest and the highest feature values can improve the performance of the algorithm VFI 5.