A case study employing non-parametric regression to develop a novel triglycerides model based on HDL levels using non-normally distributed data

Abstract

Background This research models high-density lipoprotein (HDL) and triglyceride consumption using non-parametric regression, bootstrap resampling, and data splitting into training and testing sets to improve comprehension of their complex connection and inform evidence-based health promotion efforts. Objective The goal of this study is to develop a non-parametric regression model that connects HDL levels with triglyceride levels. This will help make predictions more accurate for HDL levels in the patients that were studied by using diagnostic tools. Materials and Methods When the assumption of linear regression is not satisfied, the model that is constructed may produce biased estimates. To get around this problem, a non-parametric regression model is used in this study, along with an improved bootstrap method to get the best model estimates. In particular, the study uses the Kendall- Theil Sen Siegel slope, a robust estimator, to find the slope of the regression line. This reduces the effect of outliers or values that are too high or too low. This method makes it easy to explore the relationship between variables without having to make rigid assumptions about the parameters. This method was used to split the dataset into training and testing groups. The training dataset will be used to build the model, and the testing dataset will be used to make sure the model works. Result Based on the statistical analysis conducted using R, it was found that the non-parametric regression method performed superiorly in making predictions, particularly in instances where the data didn’t adhere to the assumption of normality. Notably, both the training and testing datasets yielded high R-squared values of 79.2% and 73.6% respectively under the non-parametric regression model. In contrast, when employing simple parametric regression, the R-squared values for the training and testing datasets were 45.89% and 47.29% respectively. A significantly high level of performance was attained using the proposed methodology, as demonstrated by these results. Conclusion The outcome of the research underscores the exceptional performance exhibited by the hybrid model methodology utilized.

BJMS, Vol. 25 No. 02 April’26 Page: 566-575