A Classification Methodology Using Linear Programming Model

Abstract

In this paper, we propose an improved method for classification that employs a combination of multiple linear programming model instances. We refer to a linear program model and confirm the linear program a problem in process of classifying a dataset, so we present how to handle such a situation. Each linear programming instance minimizes the error of the misclassified points yielding a hyperplane that classifies the dataset. Most of the existing machine learning models are based on the ‘black box’ model where the users cannot obtain any explanation regarding how the output is generated, but our approach has the potential to interpret how the output is generated because we can see the process. Furthermore, several guidelines to avoid overfitting in training procedure are provided together. We also present some experiments to confirm our method performs as efficiently as the other classification models do with various datasets.