Title Page

ABSTRACT

Contents

Chapter 1. INTRODUCTION 11

1.1. Research background 11

1.2. Objectives 13

Chapter 2. LITERATURE REVIEW 15

Chapter 3. MATERIALS AND METHODS 20

3.1. Design of experiment 20

3.1.1. Experimental environment 20

3.1.2. Experimental procedure 21

3.1.3. Experimental participants 23

3.1.4. Data collection and preprocessing 24

3.2. Time-series clustering 26

3.2.1. k-means clustering 26

3.2.2. Methods for determining the optimal number of clusters 27

3.3. Statistical analysis 29

3.4. Classification model 29

3.4.1. Classification algorithms 32

3.4.2. Forecast performance evaluation metrics 35

Chapter 4. RESULTS AND DISCUSSION 38

4.1. Time-series clustering for distinguishing high-risk group vulnerable to extremely hot environment 38

4.2. Distributions of personal biometric characteristics by the level of vulnerability 41

4.3. Development of a classification model for identifying high-risk group 47

4.4. Re-evaluation of classification model performance following outlier handling using winsorization 50

4.5. Discussion 53

Chapter 5. CONCLUSIONS 55

REFERENCES 57

APPENDICES 70

APP 1. Clustering code 70

APP 2. Classification model code 76

국문초록 106

Table 2.1. Previous studies related to workers' heat stress from the perspective of heat stress forecast 18

Table 2.2. Previous studies related to identifying individuals vulnerable to extremely hot environments 19

Table 3.1. Designed experimental conditions and measured data range 21

Table 3.2. Statistics for participants' personal characteristics 24

Table 3.3. Classification algorithms used in the previous studies on classifying specific health conditions 31

Table 4.1. Independent Samples t-test of personal biometric characteristics under extremely hot environment 43

Table 4.2. Group Statistics of personal biometric characteristics under extremely hot environment 46

Table 4.3. Forecast performance of classification models with principal features 49

Table 4.4. Independent Samples t-test of personal biometric characteristics under extremely hot environment when applying Winsorization method (for outlier handling) 51

Table 4.5. Group Statistics of personal biometric characteristics under extremely hot environment when applying Winsorization method (for outlier handling) 52

Table 4.6. Forecast performance of classification models with principal features when applying Winsorization method (for outlier handling) 53

Figure 3.1. Experimental procedure 23

Figure 3.2. Instrument information 26

Figure 3.3. Confusion Matrix 37

Figure 4.1. The best k value in the k-means clustering under extremely hot environment 39

Figure 4.2. Cluster-based time-series graphs under extremely hot environment 40

Figure 4.3. Cluster-based representative metabolic rates estimated by real-time heart rate under extremely hot environment 41

Figure 4.4. Distribution charts of personal biometric characteristics under extremely hot environment (indicators representing standard physique range) 44

Figure 4.5. Distribution charts of personal biometric characteristics under extremely hot environment (remaining eight indicators) 45

Figure 4.6. AUROC and AUPRC for six classification models 50