The  Davies–Bouldin index (DBI) is a metric for evaluating  clustering algorithms. [1]
Let  Ci be a cluster of vectors. Let  Xj be a vector on  Ci.
S_i = \sqrt[q]{ \frac{1}{T_i} \displaystyle\sum_{j=1}^{T_i}\left|X_j-A_i\right|^q}
Here  Ai is the centroid of  Ci.  Si is a measure of scatter within the cluster.
M_{i,j} = \sqrt[p]{\displaystyle\sum_{k=1}^{N}\left|a_{k,i}-a_{k,j}\right|^p }
Mi,j is a measure of separation between cluster  Ci and cluster  Cj.
ak,i is the  kth element of  Ai.
R_{i,j} \equiv \frac{S_i+S_j}{M_{i,j}}
R_i \equiv \max_{j : i \neq j} R_{i,j}
\bar{R} \equiv \frac{1}{N}\displaystyle\sum_{i=1}^N R_i

[edit]Notes and references

  1. ^ Davies, D. L.; Bouldin, D. W. A cluster separation measure.IEEE Trans. Pattern Anal. Mach. Intelligence 1979, 1, 224–227. 论文见附件