Logistic Regression Methods Versus Machine Learning Techniques In Status and Severity Prediction of South African Covid-19 Laboratory

Aug 4, 2026

Upload by Ezra

A research dissertation investigates predictive modeling for South African Covid-19 laboratory outcomes, focusing on status and severity prediction. The study compares logistic regression approaches with machine learning techniques, including random forest and self-normalising neural networks, alongside feature selection and confounder handling. Data are prepared through cleaning, imputation, and exploratory analysis, with dimensionality reduction and factor analysis to structure variables. Model performance is evaluated using appropriate measures, supporting a comparative discussion of statistical and machine learning effectiveness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages172

Reading Time7 h 13 min

AccessFree to download

Introduction

Background

Research Question

Aims and Objectives

Literature Review

Blood Parameters Identified

Theoretical Review

Imputation and Feature Selection

Logistic Regression

Artificial Neural Networks

Random Forest Modelling

Model Performance Measures

Methodology

Data Description

Data Preparation and Performance Evaluation

Exploratory Data Analysis

Machine Learning Methods

Logistic Methods

Analysis

Cleaning and Preparation of the Data

Imputation

Exploratory Data Analysis

Variable Selection

Final Datasets

Machine Learning Models

Logistic Regression Models

Summary and Discussion

Conclusion, Limitations and Recommendations

MSc Dissertation Logistic Regression Methods Versus Machine Learning Techniques In Status and Severity Prediction of South African Covid-19 Laboratory Test Data By Mark Strickett Student Number 1849246 Supervisors Dr. Charles Chimedza Dr. Farai Mlambo A research report submitted to the Faculty of Science, University of the Witwatersrand, in partial fulﬁlment of the requirements for the degree of Master of Science Contents 1 Introduction 9 1.1 Introduction ............................... 9 1.2 Background ............................... 10 1.3 Research Question ........................... 11 1.4 Aims and Objectives .......................... 11 1.4.1 Aims .............................. 11 1.4.2 Objectives ............................ 12 2 Literature Review 13 2.1 Introduction ............................... 13 2.2 Literature Review ............................ 13 2.3 Blood Parameters Identiﬁed ...................... 19 2.4 Theoretical Review ........................... 20 2.4.1 Imputation: Predictive Mean Matching Imputation ...... 20 2.4.2 Multicollinearity ........................ 20 2.4.3 Confounding Variables ..................... 21 2.4.4 Boruta Feature Selection .................... 21 2.4.5 Logistic Regression ....................... 22 2.4.5.1 Multinomial Logistic Regression .......... 22 2.4.5.2 Baseline-Category Logit Model ........... 23 2.4.5.3 Ordinal Logit Model ................. 23 2.4.6 Artiﬁcial Neural Networks ................... 24 2.4.7 Random Forest Modelling ................... 26 2.4.8 Model Performance Measures ................. 28 3 Methodology 32 3.1 Introduction ............................... 32 3.2 Research Outline ............................ 32 3.3 Data Description ............................ 34 3.4 Data Preparation and Performance Evaluation ............. 34 3.5 Exploratory Data Analysis ....................... 35 3.6 Machine Learning Methods ....................... 35 3.6.1 Random Forest Modelling ................... 36 3.6.2 Self-Normalising Neural Network (SNN) ........... 36 3.7 Logistic Methods ............................ 36 3.7.1 Multinomial Logit Model ................... 37 3.7.2 Baseline-Category Logit Model ................ 37 3.7.3 Ordinal logit Model ...................... 37 4 Analysis 39 4.1 Introduction ............................... 39 4.2 Cleaning and Preparation of the Data .................. 39 4.2.1 Multiple Dataset Creation ................... 40 4.3 Imputation ................................ 41 4.4 Exploratory Data Analysis ....................... 43 4.4.1 Descriptive Statistics ...................... 43 4.4.2 Confounding Variable Analysis ................ 53 4.4.3 Principal Component Analysis ................. 54 4.4.3.1 PCA Dataset 1 .................... 54 4.4.3.2 PCA Dataset 2 .................... 56 4.4.3.3 PCA Dataset 3 .................... 58 4.4.3.4 PCA Dataset 4 .................... 60 4.4.4 Factor Analysis ......................... 62 4.4.4.1 FA Dataset 1 ..................... 62 4.4.4.2 FA Dataset 2 ..................... 66 4.4.4.3 FA Dataset 3 ..................... 68 4.4.4.4 FA Dataset 4 ..................... 71 4.5 Variable Selection ............................ 73 4.5.1 Univariate Test ......................... 74 4.5.2 Boruta Feature Selection .................... 75 4.6 Final Datasets .............................. 79 4.7 Machine Learning Models ....................... 79 4.7.1 Random Forest Modelling ................... 79 4.7.1.1 Random Forest Model 1 ............... 80 4.7.1.2 Random Forest Model 2 ............... 82 4.7.1.3 Random Forest Model 3 ............... 84 4.7.1.4 Random Forest Model 4 ............... 87 4.7.2 Neural Networks ........................ 89 4.7.2.1 Self-normalising Neural Network 1 ......... 89 4.7.2.2 Self-normalising Neural Network 2 ......... 92 4.7.2.3 Self-normalising

FAQs

What is the main research focus of the dissertation?

The dissertation focuses on predicting Covid-19 status and severity using South African laboratory data, comparing logistic regression methods with machine learning approaches.

Which modeling techniques are compared?

The work compares multinomial, baseline-category, and ordinal logistic regression models with machine learning methods such as random forest and self-normalising neural networks.

How is the dataset prepared before modeling?

The dataset is cleaned and prepared, multiple datasets are created, missing values are handled via imputation, and exploratory data analysis is performed using descriptive statistics, confounder analysis, PCA, and factor analysis.

Aug 4, 2026

Upload by Ezra