HS-measure : a hybrid clustering validity measure to interpret road traffic data

Clustering validity measures aim to evaluate the goodness of clustering results in order to find the best partition. Results are obtained by varying the input parameters values. However, sometimes, the values generated by these measures are very close and the choice of the optimal value associated to the best partition may be meaningless. In this paper, we propose a new concept called hybrid strategy to resolve this problem. This concept is based on the use of two measures. The first measure aims to analyse the goodness of each partition obtained with different values of input parameters. The use of the second measure permits to select the best partition between those having good but very close values of the first measure. To illustrate this strategy, we propose a new hybrid measure ---called ``HS-measure"--- based on Homogeneity degree and Silhouette coefficient. The performance of our measure is then tested on road traffic data set.