Model Evaluation for Logistic Regression and Support Vector Machines in Diabetes Problem

Abstract

Machine learning is a method or computational algorithm to solve problems based on data that already available from the database. Classification is one of the important methods of supervised learning in machine learning. Support Vector Machine and Logistic Regression are some supervised learning methods that can be used both for classification and regression. In datamining process, Preprocessing is an important part before doing further analysis. In preprocessing data, feature selection and deviding training and testing data are important part of preprocessing data. In this research will be compared some evaluation model of deviding method for training and testing data, namely Random Repeated Holdout, Stratified Repeated Holdout, Random Cross-Validation, and Startified Cross-Validation. Evaluation model would be implying in logistic regression and Support Vector Machines (SVMs). From the analysis, can be concluded that by selecting features can improve the accuracy of classification with logistic regression, but opposite of Support Vector Machines (SVMs). For training and testing data pertition method can not be sure what method is better, because each method of partition training and testing data using the concept of random selection. Model evaluation cannot sure influence to increase best perform for SVMs model in particular this case.