(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Classification model to assess risk of type II diabetes based on non-intrusive parameters

Classification model to assess risk of type II diabetes based on non-intrusive parameters / Kong Xin Ping
Penyakit kencing manis semakin meningkat setiap tahun dan kematian masih tinggi. Banyak kajian mengenai klasifikasi penyakit kencing manis telah dilengkapkan tetapi kebanyakannya berdasarkan kepada ciri-ciri mengganggu. Projek ini membentangkan cara alternatif untuk menilai risiko mendapat penyakit kencing manis menggunakan ciri-ciri tidak mengganggu. Potensi ciri-ciri tidak mengganggu untuk penyakit kencing manis dikenal pasti melalui kaedah ciri-ciri pengurangan yang berbeza iaitu analisis statistik (ujian-t dan ujian Khi Kuasa Dua) dan analisis komponen utama (PCA). Ciri-ciri tidak mengganggu yang dikenalpasti digunakan untuk membangunkan ANN pengasing. Prestasi ANN daripada kedua-dua kaedah pengurangan ciri (PCA dan ujian statistik) dinilai dan dibandingkan dengan teknik pengesahan 10 kali ganda. Pengasing dengan kaedah ciri-ciri pengurangan yang paling optimum kemudiannya dibandingkan dengan penyelidik sebelum yang menggunakan ciri-ciri mengganggu. Umur, BMI, tekanan darah dan nisbah pinggang-ke-pinggul telah diiktiraf sebagai ciri-ciri tidak mengganggu yang ketara. ANN pengasing yang menggunakan teknik PCA telah mencapai ketepatan (92.9%) yang lebih baik daripada pengasing dengan teknik analisis statistik (92.5%). Prestasi model klasifikasi dengan teknik PCA boleh dibandingkan dengan kajian yang menggunakan ciri-ciri mengganggu sebelum ini yang mempunyai ketepatan dari 73.7% kepada 98.0%. Pendek kata, kajian ini telah menunjukkan bahawa pengasing penyakit kencing manis tanpa ciri-ciri mengganggu berpotensi untuk dibina. _______________________________________________________________________________________________________ Diabetes disease has been on the rise every year and the fatality remains high. Many researches about the classification of diabetes disease have been done, mostly based on intrusive features. This project presents an alternative way of assessing risk of having type II diabetes disease using non-intrusive features. Potential of non-intrusive features for diabetes disease are identified via different features reduction methods which are statistical (t-test and chi-square test) and principal component analysis (PCA). Identified non-intrusive features are used to develop ANN classifier. The performance of ANN from both feature reduction method (PCA and statistical test) are evaluated and compared using 10-fold validation technique. Classifier with the most optimum features reduction method is then compared with the previous intrusive works. BMI, blood pressure and weight to hip ratio have been recognized as significant non-intrusive features. ANN classifier using PCA technique has achieved better mean accuracy (92.9%) than classifier with statistical analysis technique (92.5%). The performance of the classification model with PCA technique is comparable with the previous intrusive works which have accuracies ranging from 73.7% to 98.0%. In short, research has shown that there is a potential of building diabetes classifier without intrusive features.
Contributor(s):
Kong Xin Ping - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875006050
Language:
English
Subject Keywords:
Diabetes; disease; intrusive features
First presented to the public:
6/1/2016
Original Publication Date:
7/4/2018
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 102
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2018-07-04 10:49:58.196
Date Last Updated
2019-01-07 11:24:32.9118
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
Classification model to assess risk of type II diabetes based on non-intrusive parameters1 2018-07-04 10:49:58.196