Skip to main content

Table 5 Clustering measures

From: Temperature variability analysis using wavelets and multiscale entropy in patients with systemic inflammatory response syndrome, sepsis, and septic shock

  Cluster points Intraclass distance Interclass distance     
  n1 n2 c1 c2 d12 d21 Clustering cost Sensitivity Specificity Accuracy
DWT high ultradian entropy (WEn5-6) 15 7 8.73 6.73 3.85 3.86 1.673 82.35% 80% 81.81%
CWT entropy neuro meta 5 17 6.15 6.79 4.25 4.24 1.749 60% 88.23% 81.81%
CWT
entropies
8 14 10.61 4.11 3.61 2.79 1.777 100% 82.35% 86.36%
CWT entro ultradian 14 8 3.95 10.76 2.68 3.57 1.788 82.35% 100% 86.36%
SampEn and sumEn 17 5 8.43 6.34 3.28 3.83 1.905 76.47% 20% 63.63%
CWT entro all and neurogenic 5 17 6.78 8.73 4.21 4.09 1.990 60% 88.23% 81.81%
DWT low ultradian entropy (WEn7-8) 15 7 13.49 9.46 3.17 3.44 2.402 76.47% 60% 72.72%
DWT Neurogenic and metabolic entropy (WEn1-2-3) 6 16 9.63 11.71 5.83 6.36 2.420 60% 82.35% 77.27%
T Mean & Std 9 13 11.63 12.78 2.35 2.13 2.499 40% 58.82% 54.54%
CWT energy 4 18 14.48 11.92 8.17 6.35 4.353 0 76.47% 59.09%
  1. The table shows, for sign_mdetr signal, clusters formed for different feature sets, number of data per cluster (n1 and n2), along with the intraclass euclidean distance for each cluster (c1 and c2), the mean interclass distance to center of the class (d12 and d21), the cost, calculated as the sum of intraclass and inverse interclass distances, and the correspondence of clustering with the a posteriori knowledge of groups (as accuracy, sensitivity, and specificity, with sensitivity referring to SIRS). The schemes are sorted in terms of cost. Schemes resulting in clustering with fewer than three members are omitted. DWT, discrete wavelet transformation; CWT, continuous wavelet transformation; SampEn, sample entropy; sumEn, multiscale entropy; SampEn, sample entropy; T, temperature; Std, standard deviation.