A methodology building for multilayer feed-forward neural network (MLFFNN): an application in biometry modelling

• Main Author: Adnan, Mohamad Nasarudin
• Format: Thesis
• Language: English
• Published: 2023
• Subjects: QH Natural history
• Online Access: http://eprints.usm.my/58898/1/MOHAMAD%20NASARUDIN%20BIN%20ADNAN-FINAL%20TESIS%20P-SGM000822%28R%29%20-24%20pages.pdf; http://eprints.usm.my/58898/

This research aims to develop a hybrid method for Multi-Layer Feed-Forward Neural Network (MLFFNN) with two different approaches; (i) Multiple Logistic Regression (MLogisticR) for the first method, (ii) Multiple Linear Regression (MLinearR) for the second method. The developed hybrid method is based on bootstrap, regression, and MLFFNN. In the first method, the accuracy of the developed method is measured based on the value of the Mean Squared Error Neural Network (MSE.net), Mean Absolute Deviance (MAD), and the accuracy percentage. While for the second method, Mean Squared Error Neural Network (MSE.net) and R2 will be used to evaluate the performance of the proposed method. All those components serve as a yardstick to determine the accuracy and efficiency of the developed model. Existing software only produces limited results. The main focus of this study is the need for better decision-making with solid evidence. The main goal of this research is to build a hybrid method and generate a numerical result and visualization (graphical representation). The results from both case studies show that the hybrid method has successfully improved the accuracy, effectiveness, and efficiency of parameter estimation in the final results of the analysis. The findings of this study contribute to the development of a comprehensive research methodology in future and suggest more accurate results for the decision-making process.