Performansi Piecewise Polynomial Smooth Support Vector Machine untuk Klasifikasi Desa Tertinggal di Provinsi Kalimantan Timur Tahun 2011

Abstract

One of most popular techniques of binary data classification in machine learning is Support Vector Machine (SVM). SVM can be applied extensively in many fields, such as pattern recognition, regression analysis, and probability estimation. SVM uses optimization with quadratic programming which become unefficient when applied in a high dimensional large dataset. Hence, researchers develop a method by changing SVM formulation with a smoothing technique that called Smooth-SVM (SSVM) which converts quadratic into linear programming. The research then continued by modifying that smooth function into polynomial smooth function forms, such as quadratic polynomial function, fourth polynomial function, piecewise polynomial function and spline function. Compared to the other polynomial smooth functions, piecewise polynomial smooth function has a better performance in plus function. When piecewise polynomial smooth function is applied in SSVM model, it will produce piecewise polynomial smooth support vector machine (PPSSVM). PPSSVM has many advantages compared to other SSVM models and its developments such as better efficiency, precision and higher accuracy in generalization. Two PPSSVM model based on piecewise polynomial function are used in this research which found by Luo (PPSSVM1) and Wu and Wang (PPSSVM2). The performance and the convergence of both models then will be examined theoretically, in order to determine the best model for classification of underdeveloped rural in East Kalimantan.PODES data in 2011 will be used in this research. Teoritical analysis showed that PPSSVM2 has a better performance and konvergence than PPSSVM1. Based on the result of this study, PPSSVM2 is not batter than PPSSVM1. It can be seen from the accuracy and AUC values that are not significantly different.