Comparative analysis in the prediction of early-stage diabetes using multiple machine learning techniques - research

Sep 17, 2026

Comparative analysis evaluates early-stage diabetes prediction using feature selection via ReliefF and tenfold cross-validation with multiple machine learning classifiers, including random forest, support vector machines, and neural networks. The work describes the clinical motivation of identifying diabetes early by detecting patterns in blood-glucose related conditions, aiming to reduce hazards through accurate risk prediction. Experiments report the highest precision with random forest at 98.5%, followed by SVM at 96.6% and neural network at 96.2%, supporting an accurate diagnostic approach for patient stratification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

1. INTRODUCTION

1.1 Objectives of the study

2. RELATED LITERATURE

2.1 Related works in diabetes

Comparative analysis in the prediction of early-stage diabetes using multiple machine learning techniques Leonard Flores1,3, Rowell Marquez Hernandez2,3, Lloyd H. Macatangay2,3, Shiela Marie G. Garcia3, Jonnah R. Melo3 1College of Arts, Sciences and Technology, Occidental Mindoro State College, San Jose, Philippines 2National Research Council of the Philippines, Division of Engineering and Industrial Research (VII) , Batangas City, Philippines 3College of Informatics and Computing Sciences-Batangas State University, Batangas City, Philippines Article history: Received Dec 18, 2021 Revised Jul 18, 2023 Accepted Jul 31, 2023 Keywords: Comparative analysis Diabetes prediction Feature selection Neural network Random forest Support vector machines Corresponding Author: Diabetes is caused by high levels from blood glucose and it is characterized as a chronic disease, and also will disrupt fat and protein absorption. The levels rise from blood glucose because it cannot be burned in the cells from the pancreas because of the deficiency of insulin secretion or the insulin produce by the cell are insufficient. By means of early detection, it may decrease the hazards and frequency of diabetes. The application of technology has been an essential part of providing accurate and acceptable results in the prevention and early detection of the illness. This research provided the best machine learning used for predicting the early stage of diabetes. The methods involve the feature selection or dimension reduction using relief-based filter (reliefF), tenfold cross-validation for testing and training data, and different machine learning classifiers such as the random forest (RF), support vector machines (SVM), and neural network (NN) is used. In this research, RF recorded the highest precision point at 98.5%, which was able to provide a higher evaluation in terms of accuracy followed by SVM at 96.6% and NNat 96.2% . The results generated from this experiment are essential in contributing a new way that is highly accurate in determining diabetes among patients. This is an open access article under the CC BY-SA license. Rowell Marquez Hernandez College of Informatics and Computing Sciences-Batangas State University Batangas City, Philippines Email: [rowell.hernandez@g.batstate-u.edu.ph](rowell.hernandez@g.batstate-u.edu.ph) Article Info ABSTRACT 1. INTRODUCTION Diabetes is caused by high levels from blood glucose and it is characterized as a chronic disease, and also will disrupt fat and protein absorption. The levels rise from blood glucose because it cannot be burned in the cells from the pancreas because of the deficiency of insulin secretion or the insulin produce by the cell are insufficient [1] . Many people are not fully aware of having early-stage diabetes symptoms, it causes frequent urination, intensifies thirst and hunger. The factors in having a diabetes such as weight, genes, and insulin, but the most of time is sugar absorption, and for preventing diabetes the better way is to detect early as possible. To determine the early stage of diabetes by providing correct classifier method and important significant attributes. Some proponents used machine learning classification to diagnose diseases [2] . The machine learning researchers used are random forest (RF), decision tree, naïve bayes, support vector machines (SVMs), and so on, as the result of their study the models or algorithms in machine learning are most efficient and effective in diagnosing diseases [3]–[5] . Machine learning and data mining handle large data from various sources and study context information are incorporated [6]–[8] . Objectives ofthe study: − To train and test data in different machine learning models and compare their performance using the earlystage diabetes 2020 dataset in terms of accuracy, precision, confusion matrix, recall, and specificity, and to train and test the data using tenfold cross-validation to see the unse

FAQs

What feature-selection method is used in this study to improve diabetes prediction?

The study uses a relief-based filter (ReliefF) for feature selection and dimension reduction to optimize model performance.

Which machine learning classifiers are compared for early-stage diabetes prediction?

Random forest (RF), support vector machines (SVM), and neural networks (NN) are trained and tested on the early-stage diabetes dataset.

What precision results does the study report for each classifier?

Random forest achieves the highest precision at 98.5%, followed by SVM at 96.6% and neural networks at 96.2%.

Sep 17, 2026

Uploaded by Evangeline