Title Page

Abstract

Contents

List of abbreviations 11

Chapter 1. Introduction 12

1.1. Background 12

1.2. Problem statement 16

1.3. Research objective 17

1.3.1. Research questions 18

Chapter 2. Literature Review 19

Chapter 3. Methodology 24

3.1. Data description 25

3.1.1. Population 25

3.1.2. Dataset construction 27

3.1.3. Exploratory data analysis 29

3.2. Machine Learning analysis framework 30

3.2.1. Predictive models 30

3.2.2. Data pre-processing 30

3.2.3. Data splitting and Cross-validation 30

Chapter 4. Experiment results 31

4.1. Determining indicators characteristics by an exploratory data analysis (EDA) 31

4.2. Prediction of cocoa production profit by using Machine Learning models 37

4.2.1. Determining organic and conventional farming main indicator by SHAP analysis 43

4.2.2. Analysis of the change over time of organic and conventional farming main indicators 47

Chapter 5. Discussion 52

Chapter 6. Conclusions 54

References 63

국문초록 75

Appendices 76

Table 3.1. Research universe described by type of farming method and variety. 26

Table 3.2. Indicators by sustainability dimension 28

Table 2.3. Groups of variety and farming method. 29

Table 4.1. Performance of ML models based con CV test 37

Table 4.2. Performance parameters among ML models 38

Table 4.3. Model performance on organic and conventional farming profit prediction. 45

Table 4.4. Cocoa production indicators' change over time 49

Table 5.1. Correlation values for General data with profit (Y) 56

Table 5.2. Correlation values for organic cocoa farming with profit (Y) 57

Table 5.3. Correlation values for conventional cocoa farming with profit (Y) 57

Figure 3.1. Methodology framework 24

Figure 4.1. Distribution of indicators 32

Figure 4.2. Distribution of output indicator "Profit" 33

Figure 4.3. Correlation between indicators 34

Figure 4.4. Irrigated area by groups 35

Figure 4.5. Cocoa production in quantity by groups. 36

Figure 4.6. Decision Tree from Random Forest predicting the profitability of cocoa production 39

Figure 4.7. Actual vs predicted values for RF, LG and kNN based on global dataset 40

Figure 4.8. Indicators importance based on SHAP values for the global dataset 41

Figure 4.9. Indicator importance based on SHAP values for organic and conventional farms 43

Figure 4.10. Actual vs predicted values for RF, LG and kNN based on organic and conventional farms 46

Figure 4.11. Conventional farming main indicators changes over time 47

Figure 4.12. Organic farming main indicators changes over time 47

Equation 3.1. Eyendi's method Equation 29