Title Page
Contents
I. Introduction 10
II. Material and methods 11
2.1. Data set 11
2.2. Feature selection methods 13
2.3. Machine learning methods 15
2.4. Validation method 17
2.5. Performance metrics 17
III. Results and discussion 19
3.1. Results of models 19
IV. Conclusion 43
References 45
Abstract 46
국문요약 47
Table 1. Number of data in each dataset. 12
Table 2. Number of features in each dataset. 13
Table 3. Statistical performance of the model with the highest F₁-score among XGBoost models with RFE applied to each filter method on CCRF-CEM dataset. 41
Figure 1. Cross validation performance of each feature selection method in naïve bayes classification model on CCRF-CEM dataset 22
Figure 2. Test set performance of each feature selection method in naïve bayes classification model on CCRF-CEM dataset 23
Figure 3. Cross validation performance of each feature selection method in random forest classification model on CCRF-CEM dataset 24
Figure 4. Test set performance of each feature selection method in random forest classification model on CCRF-CEM dataset 25
Figure 5. Cross validation performance of each feature selection method in XGBoost classification model on CCRF-CEM dataset 26
Figure 6. Test set performance of each feature selection method in XGboost classification model on CCRF-CEM dataset 27
Figure 7. Cross validation performance of each feature selection method in naïve bayes classification model on MOLT-4 dataset 28
Figure 8. Test set performance of each feature selection method in naïve bayes classification model on MOLT-4 dataset 29
Figure 9. Cross validation performance of each feature selection method in random forest classification model on MOLT-4 dataset 30
Figure 10. Test set performance of each feature selection method in random forest classification model on MOLT-4 dataset 31
Figure 11. Cross validation performance of each feature selection method in XGBoost classification model on MOLT-4 dataset 32
Figure 12. Test set performance of each feature selection method in XGboost classification model on MOLT-4 dataset 33
Figure 13. Cross validation performance of each feature selection method in naïve bayes classification model on RPMI-8226 dataset 34
Figure 14. Test set performance of each feature selection method in naïve bayes classification model on RPMI-8226 dataset 35
Figure 15. Cross validation performance of each feature selection method in random forest classification model on RPMI-8226 dataset 36
Figure 16. Test set performance of each feature selection method in random forest classification model on RPMI-8226 dataset 37
Figure 17. Cross validation performance of each feature selection method in XGBoost classification model on RPMI-8226 dataset 38
Figure 18. Test set performance of each feature selection method in XGboost classification model on RPMI-8226 dataset 39
Figure 19. Results between the feature count and F₁-score of the XGBoost model using only the filter method and the XGboost model that combines RFE with each filter method on CCRF-CEM dataset. The model with the best F₁-score... 40
Figure 20. The 20 features with the highest SHAP feature importance in the Relief-RFE-XGBoost model on CCRF-CEM dataset 42