표제지
초록
ABSTRACT
목차
기호 설명 19
1장 서론 21
2장 연구 방법 24
2.1. 모형 24
2.1.1. 콕스 비례위험 모형 24
2.1.2. Cox proportional hazards deep neural network (DeepSurv) 모형 25
2.1.3. Random survival forest 모형 27
2.1.4. Survival gradient boosting decision tree (SurvXGBoost) 모형 30
2.1.5. 시간 가변(time-varying) 앙상블 모형 32
2.1.6. 스태킹 앙상블(Stacking ensemble) 모형 38
2.2. 모형 평가 42
2.2.1. Time-dependent Brier score 42
2.2.2. Time-dependent AUROC 43
2.2.3. c-index 48
2.2.4. Greenwood-Nam-D'Agostino (GND) test 48
2.3. 사례 연구 51
2.3.1. 사례 자료 설명 51
2.3.2. 사례 분석 결과 51
3장 모의실험 & 사례연구 55
3.1. 모의실험 목적 55
3.2. 모의실험 설계 55
3.3. 모의실험 결과 평가 기준 60
3.4. 모의실험 결과 60
3.4.1. 모형 성능 평가 및 비교 61
3.4.2. 모형 성능 평가 방법의 평가 및 비교 65
4장 고찰 121
5장 결론 125
참고문헌 128
Table 1. Performance of survival prediction model with Rotterdam data 54
Table 2. List of simulation scenarios to study model performance 59
Table 3. Performance of survival prediction model when Exponential survival distribution with λ=0.01 and censoring rate=10% (iteration=200) 67
Table 4. Performance of survival prediction model when Exponential survival distribution with λ=0.01 and censoring rate=20% (iteration=200) 68
Table 5. Performance of survival prediction model when Exponential survival distribution with λ=0.01 and censoring rate=30% (iteration=200) 69
Table 6. Performance of survival prediction model when Gompertz survival distribution with λ=0.01,θ=0.08 and censoring rate=10% (iteration=200) 70
Table 7. Performance of survival prediction model when Gompertz survival distribution with λ=0.01,θ=0.08 and censoring rate=20% (iteration=200) 71
Table 8. Performance of survival prediction model when Gompertz survival distribution with λ=0.01,θ=0.08 and censoring rate=30% (iteration=200) 72
Table 9. Performance of survival prediction model when Log-logistic survival distribution with θ=3,𝒦=4 and censoring rate=10% (iteration=200) 73
Table 10. Performance of survival prediction model when Log-logistic survival distribution with θ=3,𝒦=4 and censoring rate=20% (iteration=200) 74
Table 11. Performance of survival prediction model when Log-logistic survival distribution with θ=3,𝒦=4 and censoring rate=30% (iteration=200) 75
Table 12. Performance of survival prediction model when Exponential survival distribution with λ=0.01 and censoring rate=10% (iteration=500) 76
Table 13. Performance of survival prediction model when Exponential survival distribution with λ=0.01 and censoring rate=20% (iteration=500) 77
Table 14. Performance of survival prediction model when Exponential survival distribution with λ=0.01 and censoring rate=30% (iteration=500) 78
Table 15. Performance of survival prediction model when Gompertz survival distribution with λ=0.01,θ=0.08 and censoring rate=10% (iteration=500) 79
Table 16. Performance of survival prediction model when Gompertz survival distribution with λ=0.01,θ=0.08 and censoring rate=20% (iteration=500) 80
Table 17. Performance of survival prediction model when Gompertz survival distribution with λ=0.01,θ=0.08 and censoring rate=30% (iteration=500) 81
Table 18. Performance of survival prediction model when Log-logistic survival distribution with θ=3,𝒦=4 and censoring rate=10% (iteration=500) 82
Table 19. Performance of survival prediction model when Log-logistic survival distribution with θ=3,𝒦=4 and censoring rate=20% (iteration=500) 83
Table 20. Performance of survival prediction model when Log-logistic survival distribution with θ=3,𝒦=4 and censoring rate=30% (iteration=500) 84
Figure 1. Diagram of DeepSurv 26
Figure 2. Diagram of splitting Data 39
Figure 3. Staking Ensemble 39
Figure 4. Diagram of Stacking Ensemble using random survival forest and DeepSurv 40
Figure 5. Diagram of Stacking Ensemble using SurvXGBoost and DeepSurv 41
Figure 6. Diagram of Stacking Ensemble using random survival forest and SurvXGBoost and DeepSurv 41