Risk-factor based diagnosis for chronic periodontitis using machine learning models

Name: Htun, Teza, 1996-

MeSH: Mouth -- microbiology

MeSH: Mouth Diseases

Abstract: Chronic periodontitis is one of the most common oral diseases in the world affecting 11.2% globally and 26% among Thai adults. Symptoms are negligible until it is too late and results in loss of tooth and quality of life. To diagnose chronic periodontitis, a chairside examination by a dentist or an oral hygienist is required. The process is time and resource-consuming, so a screening model to identify the risk of having chronic periodontitis in an examinee can be of assistance in reducing workload for the examiners. Cross-sectional regression models are commonly applied using relevant demographic or risk behaviors as predictors. While logistic regression models are simple to apply or to interpret, their performance can be less optimal depending on feature selection and engineering. Machine learning models recently have been increasingly applied in medical and health-related fields due to their more complex yet powerful performances and their ability to handle high dimensional data as well as unstructured data. In this study, screening models were applied such as mixed-effects logistic regression (MELR), recurrent neural networks (RNN), and mixed-effects support vector machine (MESVM). Using the Electric Generation Authority of Thailand (EGAT) cohort 2nd survey, the models were trained upon longitudinal data. Hyperparameters optimization is done for RNN and MESVM applying random-search followed by grid-search procedures. All models are evaluated with the same metrics; sensitivity, specificity, accuracy, positive likelihood ratio, positive prevalence value, negative prevalence value, Brier score, and F1 score to address class imbalance problems. It was observed that the MELR model (90.5% accuracy) performed better than the machine learning models (70.0% accuracy for RNN and 72.7% for MESVM). IMPLICATIONS OF THE THESIS. Trained models could be applied in large-scale screening in a community such as public health missions as well as electronic health records after the models have been validated with external datasets to acceptable performances.

Mahidol University. Mahidol University Library and Knowledge Center

Address: NAKHONPATHOM

Email: liwww@mahidol.ac.th

Name: Anuchate Pattanateepapon

Role: Thesis Advisors

Name: Ammarin Thakkinstian

Role: Thesis Advisors

Name: Ratchainant Thammasudjarit

Role: Thesis Advisors

Created: 2021

Modified: 2025-06-04

Issued: 2025-06-04

วิทยานิพนธ์/Thesis

application/pdf

eng

DegreeName: Master of Science

Level: Master's Degree

Descipline: Data Science for Health Care

Grantor: Mahidol University

©copyrights Mahidol University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	6238135.pdf	6.2 MB	1	2025-09-25 21:10:05

ใช้เวลา

0.44184 วินาที