Development of artificial neural network models for predicting lipid profile using smartMF electrical parameters / Ahmad Zulkhairi Zulkefli

Ahmad Zulkhairi , Zulkefli (2021) Development of artificial neural network models for predicting lipid profile using smartMF electrical parameters / Ahmad Zulkhairi Zulkefli. Masters thesis, Universiti Malaya.

	PDF (The Candidate’s Agreement) Restricted to Repository staff only Download (719Kb)
Preview	PDF (Thesis M.A) Download (1480Kb) | Preview

Abstract

This research project aims in the development of artificial neural network (ANN) to classify lipid profile parameters; total cholesterol (TC), triglycerides (TG), high density lipoprotein cholesterol (HDL-C) and low-density lipoprotein cholesterol (LDL-C) by utilizing a non-invasive bioelectrical impedance analysis (BIA) device (SmartMF). Data were obtained from volunteer patients requiring lipid profile testing at the primary care department of University Malaya Medical Centre (UMMC). The data were divided into two categories for each lipid profile parameters, normal and abnormal. Statistical analysis including T-Test and logistic regression analysis for the bioelectrical parameters were employed to understand the data and determine suitable predictors to be used for the development of the ANN. Results of the statistical analysis indicates capacitance, resistance and reactance as significant predictor for TC level. Impedance at 50, 100 and 200 kHz, resistance and reactance as significant predictors for TG level. Impedance at 5, 50, 100 and 200 kHz as significant predictors for HDL-C level. No significant predictors were determined for LDL-C level, thus ANN model for the parameter cannot be developed. ANN employing the multi-layered feed forward neural network technique was developed for the TC, TG and HDL-C parameters utilizing the scaled conjugate gradient (SCG), Levenberg Marquardt (LM) and Resilient (RB) backpropagation algorithm. The best model was then selected based on the testing performance. For TC, the SCG model with a testing accuracy of 73.3%, sensitivity of 63.6% and specificity of 78.9% was selected. For TG, the LM algorithm with a testing accuracy of 76.7%, sensitivity of 28.6% and specificity of 91.3% was selected. For HDL-C, the SCG algorithm with a testing accuracy of 76.7%, sensitivity of 95.5% and specificity of 25.0% was selected as best performing models to indicate the relationship of the bioelectrical parameters and the lipid profile parameters.

Actions (For repository staff only : Login required)