Perbandingan Algoritma Naïve Bayes dan K-Nearest Neighbor Pada Imbalace Class Dataset Penyakit Diabetes

Muhammad Rousydi Hunafa (Universitas Teknologi Yogyakarta, Yogyakarta)
Arief Hermawan (Universitas Teknologi Yogyakarta, Yogyakarta)

KLIK: Kajian Ilmiah Informatika dan Komputer
Vol. 4 No. 3 (2023): Desember 2023

Diabetes is a global health concern due to its significant impact on public health. Managing this disease is crucial to prevent serious complications. Technological advancements, particularly in machine learning models, have opened new avenues in diabetes identification. This study compares the performance of the Naive Bayes and K-Nearest Neighbor (KNN) classification algorithms on an imbalanced diabetes dataset. The primary aim is to evaluate these algorithms' performance in predicting diabetes while considering class imbalance. Classification methods were applied to previously collected datasets. The research findings demonstrate that Naive Bayes with the SMOTE technique exhibited the best performance with an accuracy of 71.66%, followed by Naive Bayes without SMOTE (76.03%), and KNN with SMOTE (80.47%). Although KNN without SMOTE showed the highest accuracy (83.02%), Naive Bayes with SMOTE showcased a better balance between accuracy, precision, and recall. The utilization of the SMOTE technique improved Naive Bayes' performance by enhancing precision and recall, indicating its capability to address class imbalance in the diabetes dataset. This study offers insights into selecting the best algorithms and effective techniques for handling class imbalance to predict diabetes on imbalanced datasets.