랜덤포레스트 기법을 활용한 청소년의 치아우식위험 예측 요인 분석
Analysis of predictive factors for dental caries risk among adolescents using the random forest algorithm
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Objectives: This study aimed to predict transitions from low/moderate to high/extreme caries risk among adolescents using the random forest (RF) algorithm and identify key contributing factors following a CAMBRA-students mobile application intervention. Methods: A quasi-experimental design with post hoc analysis was applied using data from 181 students aged 10–14 years. Of these, 23 (intervention: 15; control: 8) experienced an increase in risk. The RF model, built with 500 decision trees, was trained on pre–post changes in risk factors, seven protective factors, and four disease indicators. Class imbalance was handled using Synthetic Minority Over-sampling Technique (SMOTE), and the model performance was evaluated through cross-validation based on accuracy, precision, recall, F1 score, and area under the curve (AUC). Feature importance was assessed using permutation tests (p<0.05). Results: The RF model showed strong predictive performance (AUC: intervention=95.7%; control=99.7%). The key predictors in the intervention group included ∆R1 (frequent intake of fermentable carbohydrates), ∆D4 (tooth restoration within the past year), and ∆R4 (no use of oral hygiene items). In the control group, ∆D2, ∆D3, and ∆D4 were most important Conclusions: RF modeling effectively predicted the increase in caries risk and identified distinct predictors for each group. These f indings support the use of precision-targeted caries management.
Abstract
Objectives: This study aimed to predict transitions from low/moderate to high/extreme caries risk among adolescents using the random forest (RF) algorithm and identify key contributing factors following a CAMBRA-students mobile application intervention. Methods: A quasi-experimental design with post hoc analysis was applied using data from 181 students aged 10–14 years. Of these, 23 (intervention: 15; control: 8) experienced an increase in risk. The RF model, built with 500 decision trees, was trained on pre–post changes in risk factors, seven protective factors, and four disease indicators. Class imbalance was handled using Synthetic Minority Over-sampling Technique (SMOTE), and the model performance was evaluated through cross-validation based on accuracy, precision, recall, F1 score, and area under the curve (AUC). Feature importance was assessed using permutation tests (p<0.05). Results: The RF model showed strong predictive performance (AUC: intervention=95.7%; control=99.7%). The key predictors in the intervention group included ∆R1 (frequent intake of fermentable carbohydrates), ∆D4 (tooth restoration within the past year), and ∆R4 (no use of oral hygiene items). In the control group, ∆D2, ∆D3, and ∆D4 were most important Conclusions: RF modeling effectively predicted the increase in caries risk and identified distinct predictors for each group. These f indings support the use of precision-targeted caries management.