1 Introduction
2 Data–Observations
-
essential data was lacking (like a Cone Penetration Test);
-
a combination of installation techniques was used (both hammering and driving);
-
unexpected obstacles were expected or detected in the subsoil;
-
erroneous data was recorded, e.g. large differences (>1.5 m) between the entered length of the sheet pile and the difference between the head and the toe of the pile
-
the head of the pile was deeper than 1.5 m below the surface.
3 Prediction Models
3.1 Objective Criteria
3.2 Methods
3.2.1 Method 1 (Hypervib-I)
3.2.2 Method 2 (CUR)
3.2.3 Method 3 (Azzouzi)
3.2.4 Method 4 (Van Baars)
3.2.5 Method 5 (EAU)
3.2.6 Method 6 (Bayesian Belief Network)
4 ROC-Theory
4.1 ROC Graph and Contingency Table
Predictions | |||
---|---|---|---|
− | + | Total | |
Obs. | |||
− | TN | FP | O(−) |
+ | FN | TP | O(+) |
Total | P(−) | P(+) | N |
Predictions | |||
---|---|---|---|
− | + | Total | |
Obs. | |||
− | TN = 9 | FP = 25 | O(−) = 34 |
+ | FN = 111 | TP = 107 | O(+) = 218 |
Total | P(−) = 120 | P(+) = 132 | N = 252 |
4.2 Sensitivity Pair
4.3 Extension to ROC Curves
4.4 Conservative Predictions
Predictions | |||
---|---|---|---|
− | + | Total | |
Obs. | |||
− | TN = 30 | FP = 4 | O(−) = 34 |
+ | FN = 131 | TP = 87 | O(+) = 218 |
Total | P(−) = 161 | P(+) = 91 | N = 252 |
5 Results
5.1 Model Comparison in Current Situation
Model | Current threshold | Current sensitivity pair | Current metric distancea
| Optimized threshold | Optimized sensitivity pair | Optimized metric distancea
|
---|---|---|---|---|---|---|
CUR | 0.005 m | (0.74;0.49) | 0.90b
| 0.0029 m | (0.91;0.97) | 0.91c
|
AZZ |
F
used/F
c
>1 | (0.65;0.65) | 0.74 |
d
|
d
|
d
|
Baars |
F
used/F
c
>1 | (0.74;0.73) | 0.79 |
d
|
d
|
d
|
Hypervib-I | 0.06 m/min | (0.65;0.74) | 0.70 | 0.08 m/min | (0.44;0.62) | 0.58 |
EAU |
F
used/F
c
>1 | (0.71;0.74) | 0.76 |
d
|
d
|
d
|
BBN | 38 % | (0.59;0.74) | 0.64 | 36 % | (0.56;0.72) | 0.63 |