The evaluation of ordinal regression model's performance through the implementation of multilayer feed-forward neural network: a case study of hypertension

• Main Authors: Adnan, Mohamad Nasarudin, Ahmad, Wan Muhamad Amir W, Shahzad, Hazik, Awais, Faiza, Aleng, Nor Azlida, Noor, Nor Farid, Mohd Ibrahim, Mohamad Shafiq, M Noor, Noor Maizura
• Format: Article
• Language: English
• Published: Springer Nature 2024
• Subjects: RK Dentistry
• Online Access: http://irep.iium.edu.my/117524/1/117524_The%20evaluation%20of%20ordinal%20regression%20model%27s%20performance.pdf; http://irep.iium.edu.my/117524/; https://www.cureus.com/articles/229713-the-evaluation-of-ordinal-regression-models-performance-through-the-implementation-of-multilayer-feed-forward-neural-network-a-case-study-of-hypertension#!/; https://doi.org/10.7759/cureus.54387

Background Hypertension, or high blood pressure, is a common medical condition that affects a significant portion of the global population. It is a major risk factor for cardiovascular diseases (CVD), stroke, and kidney disorders. Objective The objective of this study is to create and validate a model that combines bootstrapping, ordered logistic regression, and multilayer feed-forward neural networks (MLFFNN) to identify and analyze the factors associated with hypertension patients who also have dyslipidemia. Material and methods A total of 33 participants were enrolled from the Hospital Universiti Sains Malaysia (USM) for this study. In this study, advanced computational statistical modeling techniques were utilized to examine the relationship between hypertension status and several potential predictors. The RStudio (Posit, Boston, MA) software and syntax were implemented to establish the relationship between hypertension status and the predictors. Results The statistical analysis showed that the developed methodology demonstrates good model fitting through the value of predicted mean square error (MSE), mean absolute deviance (MAD), and accuracy. To evaluate model fitting, the data in this study was divided into distinct training and testing datasets. The findings revealed that the results strongly support the superior predictive capability of the hybrid model technique. In this case, five variables are considered: marital status, smoking status, systolic blood pressure, fasting blood sugar levels, and high-density lipoprotein levels. It is important to note that all of them affect the hazard ratio: marital status (β1, -17.12343343; p < 0.25), smoking status (β2, 1.86069121; p < 0.25), systolic blood pressure (β3, 0.05037332; p < 0.25), fasting blood sugar (β4, -0.53880322; p < 0.25), and high-density lipoprotein (β5, 5.38065556; p < 0.25). Conclusion This research aims to develop and extensively evaluate the hybrid approach. The statistical methods employed in this study using R language show that regression modeling surpasses R-squared values in predicting the mean square error. The study's conclusion provides strong evidence for the superiority of the hybrid model technique.