1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
|
\section{Hyperparameter Sensitivity Tables}\label{app:tables}
Full numerical results for all 32 OFAT configurations ($k$: 9 values,
$\beta$: 12 values, $\gamma$: 10 combinations, $r$: 10 values).
\begin{table}[ht]
\centering
\caption{RQ3b: Sensitivity to ENM neighborhood size $k$ (50-fold CV).
Optimal at $k$=6.}
\label{tab:rq3b_new}
\footnotesize
\begin{tabular}{lcccc}
\toprule
\textbf{$k$} & \textbf{F1} & \textbf{Recall} & \textbf{Precision} & \textbf{MCC} \\
\midrule
1 & 0.848{\tiny$\pm$0.038} & 0.816{\tiny$\pm$0.047} & 0.886{\tiny$\pm$0.056} & 0.849{\tiny$\pm$0.038} \\
2 & 0.881{\tiny$\pm$0.033} & 0.847{\tiny$\pm$0.038} & 0.921{\tiny$\pm$0.058} & 0.882{\tiny$\pm$0.033} \\
3 & 0.894{\tiny$\pm$0.030} & 0.860{\tiny$\pm$0.033} & 0.933{\tiny$\pm$0.050} & 0.895{\tiny$\pm$0.030} \\
4 & 0.905{\tiny$\pm$0.035} & 0.874{\tiny$\pm$0.034} & 0.940{\tiny$\pm$0.059} & 0.906{\tiny$\pm$0.035} \\
5 & 0.908{\tiny$\pm$0.027} & 0.877{\tiny$\pm$0.036} & 0.943{\tiny$\pm$0.043} & 0.909{\tiny$\pm$0.027} \\
6 & \textbf{0.908}{\tiny$\pm$0.030} & \textbf{0.881}{\tiny$\pm$0.037} & 0.940{\tiny$\pm$0.051} & \textbf{0.909}{\tiny$\pm$0.030} \\
7 & 0.907{\tiny$\pm$0.035} & 0.879{\tiny$\pm$0.040} & 0.939{\tiny$\pm$0.055} & 0.908{\tiny$\pm$0.034} \\
8 & 0.906{\tiny$\pm$0.036} & 0.874{\tiny$\pm$0.042} & 0.944{\tiny$\pm$0.058} & 0.907{\tiny$\pm$0.036} \\
9 & 0.900{\tiny$\pm$0.036} & 0.874{\tiny$\pm$0.040} & 0.932{\tiny$\pm$0.058} & 0.901{\tiny$\pm$0.035} \\
\bottomrule
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\caption{Sensitivity to grafting temperature $\beta$ ($k$=7, 50-fold CV).}
\label{tab:rq3b_beta}
\footnotesize
\begin{tabular}{lcccc}
\toprule
\textbf{$\beta$} & \textbf{F1} & \textbf{Recall} & \textbf{Precision} & \textbf{MCC} \\
\midrule
0.5 & 0.895{\tiny$\pm$0.034} & 0.854{\tiny$\pm$0.049} & 0.942{\tiny$\pm$0.038} & 0.896{\tiny$\pm$0.034} \\
1.0 & 0.905{\tiny$\pm$0.035} & 0.871{\tiny$\pm$0.045} & 0.944{\tiny$\pm$0.043} & 0.906{\tiny$\pm$0.035} \\
1.5 & 0.908{\tiny$\pm$0.030} & 0.873{\tiny$\pm$0.042} & 0.948{\tiny$\pm$0.039} & 0.909{\tiny$\pm$0.030} \\
2.0 & 0.911{\tiny$\pm$0.030} & 0.878{\tiny$\pm$0.041} & 0.948{\tiny$\pm$0.042} & 0.912{\tiny$\pm$0.030} \\
\textbf{2.5} & \textbf{0.911}{\tiny$\pm$0.032} & \textbf{0.878}{\tiny$\pm$0.039} & 0.949{\tiny$\pm$0.044} & \textbf{0.912}{\tiny$\pm$0.032} \\
3.0 & 0.907{\tiny$\pm$0.035} & 0.879{\tiny$\pm$0.040} & 0.939{\tiny$\pm$0.055} & 0.908{\tiny$\pm$0.034} \\
3.5 & 0.906{\tiny$\pm$0.037} & 0.878{\tiny$\pm$0.039} & 0.938{\tiny$\pm$0.058} & 0.907{\tiny$\pm$0.036} \\
4.0 & 0.908{\tiny$\pm$0.034} & 0.879{\tiny$\pm$0.037} & 0.941{\tiny$\pm$0.056} & 0.908{\tiny$\pm$0.034} \\
4.5 & 0.906{\tiny$\pm$0.034} & 0.876{\tiny$\pm$0.039} & 0.940{\tiny$\pm$0.054} & 0.907{\tiny$\pm$0.034} \\
5.0 & 0.906{\tiny$\pm$0.033} & 0.878{\tiny$\pm$0.037} & 0.939{\tiny$\pm$0.056} & 0.907{\tiny$\pm$0.033} \\
5.5 & 0.905{\tiny$\pm$0.035} & 0.877{\tiny$\pm$0.041} & 0.938{\tiny$\pm$0.055} & 0.906{\tiny$\pm$0.035} \\
6.0 & 0.905{\tiny$\pm$0.036} & 0.876{\tiny$\pm$0.040} & 0.940{\tiny$\pm$0.060} & 0.906{\tiny$\pm$0.036} \\
\bottomrule
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\caption{Sensitivity to elastic range $\gamma$ ($k$=7, $\beta$=3.0, 50-fold CV).}
\label{tab:rq3b_gamma}
\footnotesize
\begin{tabular}{lcccc}
\toprule
\textbf{$\gamma$} & \textbf{F1} & \textbf{Recall} & \textbf{Precision} & \textbf{MCC} \\
\midrule
{[0.0, 0.5]} & 0.833{\tiny$\pm$0.057} & 0.769{\tiny$\pm$0.047} & 0.914{\tiny$\pm$0.094} & 0.837{\tiny$\pm$0.057} \\
{[0.0, 1.0]} & 0.865{\tiny$\pm$0.043} & 0.819{\tiny$\pm$0.049} & 0.920{\tiny$\pm$0.061} & 0.867{\tiny$\pm$0.043} \\
{[0.5, 1.0]} & 0.887{\tiny$\pm$0.032} & 0.837{\tiny$\pm$0.047} & 0.947{\tiny$\pm$0.038} & 0.889{\tiny$\pm$0.032} \\
{[0.5, 1.5]} & 0.897{\tiny$\pm$0.039} & 0.882{\tiny$\pm$0.040} & 0.915{\tiny$\pm$0.063} & 0.898{\tiny$\pm$0.039} \\
{[0.5, 2.0]} & 0.899{\tiny$\pm$0.045} & 0.888{\tiny$\pm$0.040} & 0.915{\tiny$\pm$0.076} & 0.900{\tiny$\pm$0.044} \\
{[0.8, 1.2]} & 0.906{\tiny$\pm$0.035} & 0.874{\tiny$\pm$0.044} & 0.942{\tiny$\pm$0.044} & 0.907{\tiny$\pm$0.035} \\
{[0.8, 1.5]} & 0.907{\tiny$\pm$0.035} & 0.879{\tiny$\pm$0.040} & 0.939{\tiny$\pm$0.055} & 0.908{\tiny$\pm$0.034} \\
{[0.8, 2.0]} & \textbf{0.909}{\tiny$\pm$0.035} & \textbf{0.880}{\tiny$\pm$0.042} & 0.943{\tiny$\pm$0.055} & \textbf{0.910}{\tiny$\pm$0.035} \\
{[1.0, 1.5]} & 0.906{\tiny$\pm$0.033} & 0.860{\tiny$\pm$0.044} & 0.960{\tiny$\pm$0.043} & 0.908{\tiny$\pm$0.032} \\
{[1.0, 2.0]} & 0.905{\tiny$\pm$0.037} & 0.862{\tiny$\pm$0.045} & 0.955{\tiny$\pm$0.057} & 0.906{\tiny$\pm$0.037} \\
\bottomrule
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\caption{Sensitivity to target ratio $r$ ($k$=7, $\beta$=3.0, 50-fold CV).}
\label{tab:rq3b_r}
\footnotesize
\begin{tabular}{lcccc}
\toprule
\textbf{$r$} & \textbf{F1} & \textbf{Recall} & \textbf{Precision} & \textbf{MCC} \\
\midrule
0.1 & 0.889{\tiny$\pm$0.031} & 0.826{\tiny$\pm$0.046} & 0.965{\tiny$\pm$0.044} & 0.892{\tiny$\pm$0.030} \\
0.2 & 0.900{\tiny$\pm$0.032} & 0.850{\tiny$\pm$0.044} & 0.958{\tiny$\pm$0.045} & 0.902{\tiny$\pm$0.031} \\
0.3 & 0.904{\tiny$\pm$0.029} & 0.861{\tiny$\pm$0.041} & 0.954{\tiny$\pm$0.043} & 0.905{\tiny$\pm$0.029} \\
0.4 & 0.907{\tiny$\pm$0.033} & 0.869{\tiny$\pm$0.043} & 0.951{\tiny$\pm$0.048} & 0.908{\tiny$\pm$0.032} \\
\textbf{0.5} & \textbf{0.909}{\tiny$\pm$0.031} & \textbf{0.874}{\tiny$\pm$0.039} & 0.950{\tiny$\pm$0.046} & \textbf{0.911}{\tiny$\pm$0.031} \\
0.6 & 0.907{\tiny$\pm$0.032} & 0.873{\tiny$\pm$0.041} & 0.946{\tiny$\pm$0.049} & 0.908{\tiny$\pm$0.032} \\
0.7 & 0.908{\tiny$\pm$0.031} & 0.877{\tiny$\pm$0.040} & 0.945{\tiny$\pm$0.048} & 0.909{\tiny$\pm$0.031} \\
0.8 & 0.907{\tiny$\pm$0.035} & 0.879{\tiny$\pm$0.040} & 0.939{\tiny$\pm$0.055} & 0.908{\tiny$\pm$0.034} \\
0.9 & 0.908{\tiny$\pm$0.035} & 0.877{\tiny$\pm$0.041} & 0.945{\tiny$\pm$0.052} & 0.909{\tiny$\pm$0.035} \\
1.0 & 0.909{\tiny$\pm$0.033} & 0.881{\tiny$\pm$0.040} & 0.942{\tiny$\pm$0.050} & 0.910{\tiny$\pm$0.032} \\
\bottomrule
\end{tabular}
\end{table}
\section{Factorial Interaction Results}\label{app:factorial}
Table~\ref{tab:factorial} reports the full $2^4$ factorial
experiment (16 configurations, 50-fold CV each). The maximum
two-way interaction magnitude is $0.57\%$ ($\beta\times\gamma$),
confirming the absence of significant parameter coupling.
\begin{table}[ht]
\centering
\caption{$2^4$ Factorial interaction verification (EUSES, 50-fold
CV, $n=50$ per config). Factor levels: $k\in\{5,8\}$,
$\beta\in\{2.0,4.0\}$, $\gamma\in\{(0.8,1.2),(0.8,2.0)\}$,
$r\in\{0.4,1.0\}$.}
\label{tab:factorial}
\footnotesize
\begin{tabular}{lcccc}
\toprule
\textbf{Config} & \textbf{F1} & \textbf{Recall} & \textbf{Precision} & \textbf{MCC} \\
\midrule
$k$=5 $\beta$=2.0 $\gamma$=(0.8,1.2) $r$=0.4 & 0.901$\pm$0.032 & 0.855$\pm$0.043 & 0.954$\pm$0.031 & 0.902$\pm$0.032 \\
$k$=5 $\beta$=2.0 $\gamma$=(0.8,1.2) $r$=1.0 & 0.906$\pm$0.032 & 0.864$\pm$0.044 & 0.953$\pm$0.031 & 0.907$\pm$0.031 \\
$k$=5 $\beta$=2.0 $\gamma$=(0.8,2.0) $r$=0.4 & 0.909$\pm$0.036 & 0.866$\pm$0.048 & 0.958$\pm$0.035 & 0.910$\pm$0.036 \\
$k$=5 $\beta$=2.0 $\gamma$=(0.8,2.0) $r$=1.0 & 0.908$\pm$0.036 & 0.874$\pm$0.046 & 0.946$\pm$0.041 & 0.909$\pm$0.036 \\
$k$=5 $\beta$=4.0 $\gamma$=(0.8,1.2) $r$=0.4 & 0.902$\pm$0.038 & 0.862$\pm$0.047 & 0.947$\pm$0.038 & 0.903$\pm$0.037 \\
$k$=5 $\beta$=4.0 $\gamma$=(0.8,1.2) $r$=1.0 & 0.905$\pm$0.037 & 0.871$\pm$0.046 & 0.943$\pm$0.039 & 0.906$\pm$0.037 \\
$k$=5 $\beta$=4.0 $\gamma$=(0.8,2.0) $r$=0.4 & 0.904$\pm$0.040 & 0.863$\pm$0.047 & 0.951$\pm$0.045 & 0.905$\pm$0.040 \\
$k$=5 $\beta$=4.0 $\gamma$=(0.8,2.0) $r$=1.0 & 0.903$\pm$0.042 & 0.869$\pm$0.047 & 0.941$\pm$0.052 & 0.904$\pm$0.042 \\
$k$=8 $\beta$=2.0 $\gamma$=(0.8,1.2) $r$=0.4 & 0.895$\pm$0.038 & 0.850$\pm$0.044 & 0.947$\pm$0.040 & 0.897$\pm$0.038 \\
$k$=8 $\beta$=2.0 $\gamma$=(0.8,1.2) $r$=1.0 & 0.896$\pm$0.032 & 0.856$\pm$0.039 & 0.940$\pm$0.040 & 0.897$\pm$0.032 \\
$k$=8 $\beta$=2.0 $\gamma$=(0.8,2.0) $r$=0.4 & 0.905$\pm$0.030 & 0.862$\pm$0.040 & 0.953$\pm$0.035 & 0.906$\pm$0.030 \\
$k$=8 $\beta$=2.0 $\gamma$=(0.8,2.0) $r$=1.0 & 0.906$\pm$0.031 & 0.869$\pm$0.038 & 0.948$\pm$0.042 & 0.907$\pm$0.030 \\
$k$=8 $\beta$=4.0 $\gamma$=(0.8,1.2) $r$=0.4 & 0.902$\pm$0.036 & 0.862$\pm$0.041 & 0.948$\pm$0.045 & 0.904$\pm$0.036 \\
$k$=8 $\beta$=4.0 $\gamma$=(0.8,1.2) $r$=1.0 & 0.899$\pm$0.030 & 0.867$\pm$0.039 & 0.934$\pm$0.042 & 0.899$\pm$0.030 \\
$k$=8 $\beta$=4.0 $\gamma$=(0.8,2.0) $r$=0.4 & 0.903$\pm$0.030 & 0.863$\pm$0.038 & 0.948$\pm$0.044 & 0.904$\pm$0.030 \\
$k$=8 $\beta$=4.0 $\gamma$=(0.8,2.0) $r$=1.0 & 0.905$\pm$0.032 & 0.874$\pm$0.038 & 0.941$\pm$0.045 & 0.906$\pm$0.032 \\
\bottomrule
\end{tabular}
\end{table}
\section{RQ2 Detailed Results}\label{app:rq2}
\begin{table}[ht]
\centering
\caption{RQ2 CNFS threshold sweep (EUSES fault cells, 50-fold CV).}
\label{tab:rq2_sweep}
\footnotesize
\begin{tabular}{lrrr}
\toprule
\textbf{Threshold $\tau$} & \textbf{Coverage (\%)} & \textbf{SDA F1} & \textbf{Best Baseline F1} \\
\midrule
0.0 & 100.0\% & 0.938 & 0.902 \\
0.1 & 99.6\% & 0.939 & 0.902 \\
0.2 & 99.6\% & 0.939 & 0.902 \\
0.3 & 96.4\% & 0.940 & 0.903 \\
\textbf{0.4} & \textbf{86.8\%} & \textbf{0.941} & \textbf{0.903} \\
0.5 & 71.8\% & 0.941 & 0.900 \\
0.6 & 49.0\% & 0.944 & 0.899 \\
0.7 & 28.5\% & 0.946 & 0.910 \\
0.8 & 7.0\% & 0.957 & 0.921 \\
0.9 & 3.2\% & 0.977 & 0.952 \\
\bottomrule
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\caption{Marginal contribution decomposition by CNFS bin
($n=688$ fault cells). F1 equals recall because the detection
matrix contains only fault cells. SDA-exclusive TP: cells detected
by SDA but missed by all three baselines.}
\label{tab:rq2_marginal}
\footnotesize
\begin{tabular}{lrrrrr}
\toprule
\textbf{CNFS bin} & \textbf{Cells} & \textbf{SDA F1} & \textbf{Best base F1} & \textbf{$\Delta$F1} & \textbf{SDA-excl TP} \\
\midrule
{[0.1,0.2)} & 3 & 1.000 & 1.000 & 0.000 & 0 \\
{[0.2,0.3)} & 22 & 0.955 & 0.864 & +0.091 & 0 \\
{[0.3,0.4)} & 66 & 0.894 & 0.833 & +0.061 & 4 \\
{[0.4,0.5)} & 103 & 0.816 & 0.757 & +0.058 & 5 \\
{[0.5,0.6)} & 157 & 0.898 & 0.822 & +0.076 & 7 \\
{[0.6,0.7)} & 141 & 0.894 & 0.830 & +0.064 & 9 \\
{[0.7,0.8)} & 148 & 0.892 & 0.824 & +0.068 & 4 \\
{[0.8,0.9)} & 26 & 0.885 & 0.923 & $-0.038$ & 0 \\
{[0.9,1.0]} & 22 & 0.864 & 0.818 & +0.045 & 2 \\
\midrule
All & 688 & 0.884 & 0.821 & +0.062 & 31 \\
\bottomrule
\end{tabular}
\end{table}
|
