Combination of Methodology Building for Multi-Layer FFED Forward Neural Network (MLFF) and Linear Modelling (LM): A Case Study by Biometry Modelling

Main Article Content

Muhammad Khairan Shazuan Jusoff, Wan Muhamad Amir W. Ahmad, Nor Farid Mohd. Noor, Nor AzlidaAleng, Nur FatihaGhazalli, Mohamad Shafiq Mohd. Ibrahim, Farah Muna Mohamad Ghazali,Mohamad Nasarudin Adnan, Nurfadhlina Abdul Halim

Abstract

Background: The purpose of this study is to develop and illustrate an optimum variable selection approach using Multiple Linear Regression (MLR) and validation using Multi-Layer Feed Forward Neural Network (MLFF), also known as Multilayer Perceptron Neural Network (MLP) while taking bootstrapping into account. All of the factors specified will be evaluated to determine if they have a meaningful association.


Objective:The goal of this study was to create a new method in making predictions for oral health data by combining a few statistical techniques. This would make the model more accurate.


Material and Methods:A set of medical data that consists 30 observations was used to develop the methodology. The data descriptions of variables being used in this retrospective research including Creatinine, Fasting Blood Sugar, Haemoglobin A1C, and Urea were evaluated using advanced computational statistical modelling approaches. The medical data are used to test the R syntax that is developed in this study. The statistics for each sample are calculated using a model that incorporates bootstrapping and multiple linear regression techniques.


Results:The statistical strategy which combines Bootstrap, MLFF and MLR is better than regular statistical method being used for this type of data. The hybrid model technique’s accuracy was increased when the data was separated into training and testing dataset. Four variables are taken into account in this case: fasting blood sugar, creatinine, haemoglobin A1C,  and urea. Fasting Blood Sugar ( : 0.461889; p< 0.05), Creatinine ( : 0.029761; p< 0.05) and Urea ( :-0.454766; p< 0.05) are all significant, whereas MLFF and MLRhave the predicted mean square error (PMSE) values of 0.01226 and 0.3531 respectively.


Conclusion: The value of PMSE is mainly used to diagnose the performance of MLR and MLFF. The MLFF is used to determine how close predicted values are to real data, while the model that has low PMSE value indicates that the model is accurate. The R syntax for combining Bootstrap, MLR, and MLFF is also included in this research article. In a word, this research establishes the superiority of the hybrid model method.

Article Details

Section
Articles