Development of New Hybrid Models for Prediction of Maximal Oxygen Uptake (VO2max) Using Machine Learning Methods Combined with Feature Selection Algorithms

Table of Contents

1. INTRODUCTION

1.1. VO2max as a Predictor of Fitness and Health

1.2. Measuring VO2max

1.3. Motivation in Predicting VO2max

1.4. Types of VO2max Prediction Models

1.5. Purpose and Contributions of the Thesis

1.6. Roadmap of the Thesis

2. LITERATURE REVIEW

2.1. Prediction Models Based on Exercise Tests

2.1.1. Maximal Prediction Models

2.1.2. Submaximal Prediction Models

2.2. Non-Exercise Prediction Models

2.3. Hybrid Prediction Models

3. OVERVIEW OF METHODS

3.1. Support Vector Machines

3.1.1. Linear SVM

3.1.2. Nonlinear SVM

3.2. Artificial Neural Network-Based Methods

3.2.1. Multilayer Feed-Forward Artificial Neural Network

3.2.2. General Regression Neural Network

3.2.3. Radial Basis Function Neural Network

3.3. Tree-Structured Methods

3.3.1. Tree Boost

3.3.2. Decision Tree Forest

3.3.3. Single Decision Tree

3.4. Individual Feature Selectors

3.4.1. Relief-F

3.4.2. Minimum Redundancy Maximum Relevance Feature Selector

3.4.3. Maximum-Likelihood Feature Selector

3.5. Proposed Ensemble Feature Selector

3.6. Model Validation and Testing

4. DATASET GENERATION

4.1. VO2max-set-1

4.2. VO2max-set-2

4.3. VO2max-set-3

5. DEVELOPMENT OF PREDICTION MODELS

5.1. Methodology

5.1.1. Model Creation Methodology

5.1.2. Model Evaluation Metrics

5.2. Overview and Details of Prediction Models

5.2.1. Prediction Models Created Using VO2max-set-1

5.2.2. Prediction Models Created Using VO2max-set-2

5.2.3. Prediction Models Created Using VO2max-set-3

5.2.4. SVM-Based Models for predicting VO2max

5.2.5. Artificial Neural Network-Based Models for predicting VO2max

5.2.6. Tree-Structured Models for predicting VO2max

6. RESULTS AND DISCUSSION

6.1. Results and Discussion for VO2max-set-1

6.1.1. Results for VO2max-set-1

6.1.2. Discussion on VO2max-set-1 Results

6.2. Results and Discussion for VO2max-set-2

6.2.1. Results for VO2max-set-2

6.2.2. Discussion on VO2max-set-2 Results

6.3. Results and Discussion for VO2max-set-3

6.3.1. Results for VO2max-set-3

6.3.2. Discussion on VO2max-set-3 Results

6.4. Comparing Results with Those of MVFS-Based Prediction Models

6.5. Comparing Results with Previous Works

6.6. General Discussion on Overall Results

7. CONCLUSION

Research Objectives and Themes

The primary objective of this thesis is to develop novel, accurate hybrid prediction models for maximal oxygen uptake (VO2max) by combining maximal, submaximal, and questionnaire-based physiological variables with advanced machine learning methods and feature selection algorithms. The research seeks to overcome the limitations of traditional, direct laboratory-based VO2max measurement.

• Development of hybrid feature selection models for VO2max prediction using combined variable categories.
• Implementation of seven distinct machine learning regression methods, including Support Vector Machines (SVM) and Artificial Neural Networks (ANN).
• Design and validation of a new ensemble feature selector, the Majority Voting Feature Selector (MVFS).
• Comparative analysis of prediction accuracy using multiple correlation coefficients (R) and root mean squared errors (RMSE).
• Investigation of variable importance across different populations and datasets.

Excerpt from the Book

Summary of Chapters

1. INTRODUCTION: Introduces the significance of VO2max for health and fitness and outlines the motivation for developing predictive models due to the limitations of direct measurement.

2. LITERATURE REVIEW: Provides a survey of existing VO2max prediction models based on exercise tests, non-exercise data, and hybrid approaches, comparing their performance metrics.

3. OVERVIEW OF METHODS: Details the theoretical foundations of the machine learning algorithms and feature selection methods employed, including the proposed ensemble feature selector.

4. DATASET GENERATION: Describes the composition and characteristics of the three distinct datasets (VO2max-set-1, VO2max-set-2, VO2max-set-3) used for model training and validation.

5. DEVELOPMENT OF PREDICTION MODELS: Explains the methodology for creating hybrid models, including parameter settings and the application of feature selection algorithms on the datasets.

6. RESULTS AND DISCUSSION: Presents the performance metrics (R and RMSE) of the generated models and discusses the findings regarding variable relevance and algorithm effectiveness.

7. CONCLUSION: Summarizes the study’s contributions, confirms the superior accuracy of the proposed hybrid and ensemble-based models, and suggests directions for future research.

Keywords

Machine Learning, Feature Selection, Maximal Oxygen Uptake, VO2max, Support Vector Machine, Artificial Neural Networks, Predictive Modeling, Hybrid Models, Ensemble Learning, Cardiorespiratory Fitness, Exercise Physiology, Regression Analysis, Cross-Validation.

Frequently Asked Questions

What is the primary focus of this doctoral thesis?

The thesis focuses on developing new, accurate hybrid prediction models for maximal oxygen uptake (VO2max) by combining various exercise and questionnaire variables with advanced machine learning techniques and feature selection.

What are the central themes of this research?

The central themes include the optimization of prediction accuracy, the reduction of variable redundancy through feature selection, and the benchmarking of machine learning algorithms for physiological data.

What is the primary objective of this work?

The primary objective is to create prediction models that outperform existing literature in accuracy, thereby providing a more practical and accessible alternative to costly, laboratory-based maximal graded exercise tests.

Which machine learning methods are utilized?

The study employs seven machine learning methods: Support Vector Machine (SVM), Multilayer Feed-Forward Artificial Neural Network (MFANN), General Regression Neural Networks (GRNN), Radial Basis Function Neural Network (RBFNN), Tree Boost (TB), Decision Tree Forest (DTF), and Single Decision Tree (SDT).

What does the main body of the thesis cover?

The main body covers a comprehensive literature review, the detailed methodology of the machine learning and feature selection processes, the description of three unique datasets, the development of the hybrid models, and an extensive discussion of the comparative performance results.

Which keywords characterize this work?

The key concepts include Machine Learning, Feature Selection, Maximal Oxygen Uptake, Predictive Modeling, and Hybrid Models.

What is the novelty of the Majority Voting Feature Selector (MVFS)?

The MVFS is an original ensemble feature selector that aggregates the consensus from Relief-F, mRMR, and MLFS to produce more robust and accurate prediction models by reducing the risk of selecting irrelevant predictors.

How does this study compare to the previous research of Aktürk (2014)?

This study directly compares its results with Aktürk (2014), demonstrating that the hybrid models developed here achieve significantly lower error rates by combining triple variable sets, compared to the regular models used in the previous study.