Construction and Comparison of Diabetes Prediction Models Based on GA-LightGBM.

Saved in:

Bibliographic Details
Title:	Construction and Comparison of Diabetes Prediction Models Based on GA-LightGBM.
Authors:	Liu, Xingyue¹ 918894770@qq.com, Qu, Zhenhua² 18463064907@163.com, Cao, Yuan³ yuancao@sdut.edu.cn
Source:	IAENG International Journal of Applied Mathematics. Jun2026, Vol. 56 Issue 6, p2140-2149. 10p.
Subjects:	Boosting algorithms, Genetic algorithms, Machine learning, Feature selection, Blood sugar monitoring, Medical forecasting
Abstract:	Since the dawn of the 21st century, shifts in global dietary patterns and lifestyles have precipitated a rapid surge in the incidence and prevalence of diabetes worldwide. Against this backdrop, harnessing machine learning techniques in the medical field to improve the accuracy of blood glucose detection has emerged as a critical imperative. The primary objective of this study is to develop a novel blood glucose prediction model using machine learning methodologies. By amalgamating the strengths of individual predictive models, this research aims to enhance the precision of blood glucose detection. To this end, a LightGBM model optimized with a genetic algorithm (GA-LightGBM) is proposed. The dataset utilized in this study is derived from the 2017 Tianchi Precision Medicine Competition: AI-Assisted Diabetes Genetic Risk Prediction project, comprising 5,120 samples. Each sample features 42 indicators, including unique ID, physical examination date, and gender. The dataset is randomly split into training and testing subsets at a ratio of 2:1. Feature selection is performed using the random forest algorithm, ultimately identifying ten variables that exert a significant influence on the dependent variable (blood glucose level). Subsequent to feature extraction, a genetic algorithm is employed to optimize the hyperparameters of the LightGBM model, thereby mitigating the errors associated with manual parameter tuning. The performance of the optimized model is evaluated using three key metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), and the coefficient of determination (R²). Comparative analyses reveal that the GA-LightGBM model outperforms other mainstream machine learning algorithms, exhibiting lower prediction errors and stronger interpretability. Notably, the GA-LightGBM algorithm achieves the smallest MSE among all candidate models. Beyond cross-algorithm comparisons, the genetic algorithm is also benchmarked against other prevalent hyperparameter tuning methods, demonstrating superior efficiency in terms of runtime and predictive accuracy. In conclusion, the findings of this study validate the efficacy of the GA-LightGBM algorithm, which can serve as a valuable reference and practical tool for the development of high-precision blood glucose prediction models. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Applied Mathematics is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Since the dawn of the 21st century, shifts in global dietary patterns and lifestyles have precipitated a rapid surge in the incidence and prevalence of diabetes worldwide. Against this backdrop, harnessing machine learning techniques in the medical field to improve the accuracy of blood glucose detection has emerged as a critical imperative. The primary objective of this study is to develop a novel blood glucose prediction model using machine learning methodologies. By amalgamating the strengths of individual predictive models, this research aims to enhance the precision of blood glucose detection. To this end, a LightGBM model optimized with a genetic algorithm (GA-LightGBM) is proposed. The dataset utilized in this study is derived from the 2017 Tianchi Precision Medicine Competition: AI-Assisted Diabetes Genetic Risk Prediction project, comprising 5,120 samples. Each sample features 42 indicators, including unique ID, physical examination date, and gender. The dataset is randomly split into training and testing subsets at a ratio of 2:1. Feature selection is performed using the random forest algorithm, ultimately identifying ten variables that exert a significant influence on the dependent variable (blood glucose level). Subsequent to feature extraction, a genetic algorithm is employed to optimize the hyperparameters of the LightGBM model, thereby mitigating the errors associated with manual parameter tuning. The performance of the optimized model is evaluated using three key metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), and the coefficient of determination (R²). Comparative analyses reveal that the GA-LightGBM model outperforms other mainstream machine learning algorithms, exhibiting lower prediction errors and stronger interpretability. Notably, the GA-LightGBM algorithm achieves the smallest MSE among all candidate models. Beyond cross-algorithm comparisons, the genetic algorithm is also benchmarked against other prevalent hyperparameter tuning methods, demonstrating superior efficiency in terms of runtime and predictive accuracy. In conclusion, the findings of this study validate the efficacy of the GA-LightGBM algorithm, which can serve as a valuable reference and practical tool for the development of high-precision blood glucose prediction models. [ABSTRACT FROM AUTHOR]
ISSN:	19929978