Selecting the right machine learning for community blood glucose screening in an ageing Thai population
Abstract
Community-based screening is central to managing non-communicable diseases (NCDs) in Thailand’s ageing population, yet village protocols reduce continuous glucose readings to binary risk labels. Using records from 684 participants screened in January 2025 at a community hospital in Sing Buri Province, Thailand, this study asks whether regression or classification is the safer paradigm for capillary blood glucose screening using Dextrostix (DTX). Three algorithms, multiple linear regression (MLR), random forest (RF), and eXtreme gradient boosting (XGB), were benchmarked in both paradigms across four feature-selection strategies: univariate filtering, least absolute shrinkage and selection operator (LASSO) regularization, recursive feature elimination (RFE), and full-feature inclusion. On the 20% hold-out set (n=137), the best regression model (XGB-LASSO) reached only root mean square error (RMSE)=16.0284 and coefficient of determination (R²)=0.2731, and surveillance error grid (SEG) and Bland-Altman analyses showed it was clinically unsafe, with a flattened, heteroscedastic error profile that overestimates low and underestimates high glucose. Classification proved the more reliable choice: the XGB-RFE model discriminated elevated from normal glucose at the referral threshold (accuracy=0.7226, F1=0.7146, and area under the receiver (AUC)=0.7796), and Shapley additive explanations (SHAP) analysis identified the dominant glycemic drivers, comorbid diabetes mellitus (DM) and blood pressure (BP), for transparent risk attribution. XGB-RFE gives nurses and volunteers a clear refer-or-reassure decision, supporting safer community glucose screening in low-resource settings.
Keywords
Ageing society; Gradient boosting; Non-communicable diseases; Risk stratification; Shapley additive explanations interpretation
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.13971
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.