Skip to main content
. 2022 Mar 19;94(8):821–835. doi: 10.1007/s11265-022-01748-5

Table 4.

Accelerometer-based cough detection results for the shallow classifiers. The values are averaged over 14 cross-validation folds. The highest AUC of 0.8587 has been achieved from an MLP classifier.

Classifier ID Best Feature Best Classifier Hyperparameters Performance
Hyperparameters (Optimised inside nested cross-validation) Spec Sens Acc AUC σAUC
LR C1 Ψ=16, C=5 γ1=10-4, γ2 = 0.35, γ3 = 0.65 80.41% 80.28% 80.35% 0.8055 0.003
C2 Ψ=16, C=10 γ1=10-2, γ2 = 0.55, γ3 = 0.45 80.25% 80.08% 80.16% 0.8058 0.003
C3 Ψ=32, C=5 γ1=102, γ2 = 0.2, γ3 = 0.8 80.39% 80.55% 80.47% 0.8072 0.003
C4 Ψ=32, C=10 γ1=10-3, γ2 = 0.4, γ3 = 0.6 81.42% 81.28% 81.35% 0.8135 0.003
C5 Ψ=64, C=5 γ1=10-1, γ2 = 0.25, γ3 = 0.75 80.22% 80.41% 80.31% 0.8119 0.003
C6 Ψ=64, C=10 γ1=10-2, γ2 = 0.75, γ3 = 0.25 80.16% 80.32% 80.24% 0.8124 0.003
SVM C7 Ψ=16, C=5 γ1=103, γ4=10-3 80.71% 82.91% 81.81% 0.8202 0.003
C8 Ψ=16, C=10 γ1=10-2, γ4=102 80.22% 82.94% 81.58% 0.8248 0.003
C9 Ψ=32, C=5 γ1=103, γ4=10-2 80.41% 82.97% 81.69% 0.8212 0.003
C10 Ψ=32, C=10 γ1=10-1, γ4=10-1 80.91% 84.11% 82.51% 0.8252 0.003
C11 Ψ=64, C=5 γ1=10-4, γ4=10-3 80.28% 84.35% 82.31% 0.8245 0.003
C12 Ψ=64, C=10 γ1=102, γ4=10-4 80.55% 82.78% 81.68% 0.8251 0.003
MLP C13 Ψ=16, C=5 γ3=0.55, γ5=30 82.37% 86.95% 84.68% 0.8507 0.003
C14 Ψ=16, C=10 γ3=0.45, γ5=50 83.24% 87.08% 85.16% 0.8558 0.003
C15 Ψ=32, C=5 γ3=0.35, γ5=70 83.55% 87.41% 85.47% 0.8552 0.003
C16 Ψ=32, C=10 γ3=0.4, γ5=20 82.18% 86.05% 84.12% 0.8424 0.003
C17 Ψ=64, C=5 γ3=0.7, γ5=40 84.47% 86.89% 85.67% 0.8587 0.003
C18 Ψ=64, C=10 γ3=0.35, γ5=30 83.45% 86.84% 84.64% 0.8499 0.003