Title Page
ABSTRACT
Contents
Chapter 1. Introduction 9
Chapter 2. Statistical Models 11
2.1. Keywords 11
2.1.1. Anomaly 11
2.1.2. Imbalanced data 12
2.2. Literature Review 13
2.2.1. Unsupervised Learning 13
2.2.2. Semi-Supervised Learning 17
2.2.3. Real-time Learning 18
2.3. Robust Random Cut Forest Model 21
2.3.1. Different Concepts of anomaly in models 21
2.3.2. Algorithm 22
2.4. Conformal Prediction 24
2.5. Real-time Learning model with RRCF+CAD 25
2.6. Metrics 26
Chapter 3. Case Study 28
3.1. Data Description[원문불량;p.20] 28
3.2. Data Analysis 38
3.3. Analysis Results 40
3.3.1. RRCF and RRCF+CAD 40
3.3.2. RRCF+CAD and Other Machine Learning Models 44
Chapter 4. Conclusion 48
Bibliography 50
국문초록 53
Table 3.1. Pump sensor data 29
Table 3.2. Evaluation metircs for total data : RRCF vs RRCF+CAD 44
Table 3.3. Evaluation metrics for 1th cycle : RRCF+CAD vs others 45
Table 3.4. Evaluation metrics for 2th cycle : RRCF+CAD vs others 45
Table 3.5. Evaluation metrics for 3th cycle : RRCF+CAD vs others 46
Table 3.6. Evaluation metrics for 4th cycle : RRCF+CAD vs others 46
Table 3.7. Evaluation metrics for 5th cycle : RRCF+CAD vs others 46
Table 3.8. Evaluation metrics for 6th cycle : RRCF+CAD vs others 46
Table 3.9. Evaluation metrics for 7th cycle : RRCF+CAD vs others 47
Figure 3.1. Time series plot : sensor_00 ~ sensor_07 31
Figure 3.2. Time series plot : sensor_08 ~ sensor_14, sensor_15 32
Figure 3.3. Time series plot : sensor_17 ~ sensor_24 33
Figure 3.4. Time series plot : sensor_25 ~ sensor_32 34
Figure 3.5. Time series plot : sensor_33 ~ sensor_40 35
Figure 3.6. Time series plot : sensor_41 ~ sensor_48 36
Figure 3.7. Time series plot : sensor_49, sensor_51 37
Figure 3.8. shingling example : k=3 39
Figure 3.9. failure prediction alerts when 1,2,3,4th failure 42
Figure 3.10. failure prediction alerts when 5,6,7th failure 43