Hasim As'ari Hasim, Kusrini

PENINGKATAN KINERJA KLASIFIKASI DIABETES MENGGUNAKAN METODE SUPPORT VECTOR MACHINE (SVM) DENGAN KERNEL RADIAL BASIS FUNCTION (RBF)

Introduction

Peningkatan kinerja klasifikasi diabetes menggunakan metode support vector machine (svm) dengan kernel radial basis function (rbf). Optimalkan klasifikasi diabetes dengan SVM RBF. Preprocessing data & ADASYN pada dataset Pima Indians capai akurasi 76,6% & ROC AUC 0,861. Solusi deteksi dini otomatis.

Abstract

Diabetes Mellitus merupakan salah satu penyakit kronis dengan prevalensi yang terus meningkat secara global. Deteksi dini terhadap penyakit ini sangat penting guna mengurangi risiko komplikasi yang lebih parah. Penelitian ini bertujuan untuk mengevaluasi performa algoritma Support Vector Machine (SVM) dengan kernel Radial Basis Function (RBF) dalam mengklasifikasikan penderita diabetes berdasarkan data medis. Untuk meningkatkan performa model, diterapkan berbagai tahapan preprocessing seperti normalisasi dengan StandardScaler, pembangkitan fitur non-linear dengan PolynomialFeatures, seleksi fitur dengan SelectKBest, serta penyeimbangan kelas menggunakan ADASYN. Dataset yang digunakan adalah Pima Indians Diabetes dari Kaggle, yang memiliki permasalahan ketidakseimbangan kelas. Hasil evaluasi menunjukkan bahwa model mampu mencapai nilai akurasi sebesar 76,6% dan nilai ROC AUC sebesar 0,861. Temuan ini menunjukkan bahwa pendekatan berbasis machine learning dengan pipeline yang tepat dapat menjadi solusi yang andal untuk mendukung deteksi dini Diabetes Mellitus secara otomatis.

Review

This study addresses the critical issue of Diabetes Mellitus, a chronic disease with rising global prevalence where early detection is paramount to mitigating severe complications. The authors propose an approach centered on enhancing the classification performance of Support Vector Machine (SVM) models, specifically utilizing a Radial Basis Function (RBF) kernel, for identifying diabetes patients based on medical data. The research is highly relevant given the growing need for automated and reliable diagnostic tools in healthcare, aiming to contribute a robust method for early diagnosis. To achieve improved model performance, the paper details a comprehensive machine learning pipeline. This includes crucial preprocessing steps such as data normalization using StandardScaler, generation of non-linear features via PolynomialFeatures, and targeted feature selection through SelectKBest. A notable aspect of the methodology is the explicit handling of class imbalance, a common challenge in medical datasets, which was addressed by employing the ADASYN technique. The investigation utilizes the well-known Pima Indians Diabetes dataset from Kaggle, which inherently exhibits this class imbalance problem, thus validating the chosen approach for data preparation. The evaluation of the proposed model yielded promising results, demonstrating an accuracy of 76.6% and a strong ROC AUC score of 0.861. These performance metrics suggest that the integrated pipeline, combining SVM with RBF kernel and a series of intelligent preprocessing and balancing techniques, effectively enhances diabetes classification. The findings collectively advocate for this machine learning-based approach as a robust and reliable solution, capable of significantly supporting the automated early detection of Diabetes Mellitus, thereby contributing positively to diagnostic efforts.

Full Text

Comments

You need to be logged in to post a comment.