Ponente
Descripción
We present the use of simulated annealing for clustering histogram symbolic data Billard & Diday (2020), by the minimization of a criterion based on a Huygens-type decomposition of inertia defined by Wasserstein distance. An efficient cooling scheme for simulated annealing was implemented Aarts & Korst (1990), with variable length Markov chains, allowing a large exploration in the search space. A simplification of inertia change was found in order to efficiently use Metropolis rule. The algorithm was tested on maximum daily rainfall data sets for last 40 years in Costa Rica, in 14 meteorological stations in the Reventazón river basin. Results were compared to a k-means algorithm Irpino et al. (2014), with a general improvement in quality.
References
1. Aarts, E., Korst, J.: Simulated Annealing and Boltzmann Machines. Wiley, New York (1990)
2. Billard, L., Diday, E.: Clustering Methodology for Symbolic Data. Wiley, New York (2020)
3. Irpino, A., Verde, R., De Carvalho, F.A.T.: Dynamic clustering of histogram data based on adaptive
squared Wasserstein distances. Expert Systems with Appl. 41, 3351--3366 (2014). doi:
10.1016/j.eswa.2013.12.001