Comparison of Neural Network Algorithms, Naive Bayes and Logistic Regression to predict diabetes

Abstract

Diabetes is a disease that affects many people with the characteristics of high blood sugar levels. The International Diabetic Federation (IDF) estimates the number of Indonesians aged 20 years and over, suffering from diabetes at 5.6 million people in 2001, and increasing to 8.2 million people in 2020. The problem that occurs is that many people do not know that they suffer from diabetes because they do not have basic knowledge about diabetes and the existing methods to detect diabetes are time consuming. In this study, three data mining methods were compared, namely the neural network algorithm, naïve Bayes, and logistic regression using the rapid miner application by applying the Confusion Matrix Evaluation (Accuracy) and the ROC Curve. The result of this research is that logistic regression method is a fairly good method in predicting early diagnosis of diabetes compared to the naïve Bayes method and the neural network. From the evaluation and validation, it is known that logistic regression has the highest accuracy and AUC values among the comparable methods, namely 75.78% and AUC 0.801, followed by the naïve Bayes algorithm which is 74.87% and AUC 0.799, and the neural network is 69.27% and AUC 0.736. has the lowest accuracy.