Utilizing of preprocessing PCA on VFI5 algorithm

Abstract

Nowadays, classification methods are growing rapidly. One of them is a Voting Feature Interval algorithm (VFI5). VFI5 is a supervised learning algorithm by generates a function that maps inputs to desired outputs. Phases of the training VFI5 algorithm produce the intervals. Large number of features can be effected many interval are produced, the other way little features can be effected less of interval. If data have large number of features, then the classification process is not efficiently. So, preprocessing data are needed by applying PCA for reducing feature dimension to make more effisien process. Data is used in this research, has three times repetition process using preprocessing PCA and without preprocessing PCA. Average best accuracy on Iris using preprocessing PCA is 92.67% and without PCA is 96%. Next, average best accuracy for data Wine using preprocessing PCA is 72.09% and without PCA is 93.22%, different with two previously data, New Thyroid have average best accuracy using preprocessing PCA is 91.72% and without PCA is 88.33%. Comparasion accuracy between data using preprocessing PCA and data without preprocessing PCA, only New Thyroid that using preprocessing PCA has higher accuracy than New Thyroid without preprocessing PCA. This condition caused contribution value of data New Thyroid is spread out, not gather in first principal component. Keywords : machine learning, classification, PCA, voting feature interval.