Article ID: | iaor2009864 |
Country: | Germany |
Volume: | 18 |
Issue: | 1 |
Start Page Number: | 41 |
End Page Number: | 46 |
Publication Date: | Mar 2008 |
Journal: | Statistics and Computing |
Authors: | Hansen P., Mladenovi N. |
Keywords: | heuristics |
A recent comparison of evolutionary, neural network, and scatter search heuristics for solving the p-median problem is completed by (i) gathering or obtaining exact optimal values in order to evaluate errors precisely, and (ii) including results obtained with several variants of a variable neighborhood search (VNS) heuristic. For a first, well-known, series of instances, the average errors of the evolutionary and neural network heuristics are over 10% and more than 1000 times larger than that of VNS. For a second series, this error is about 3% while the errors of the parallel VNS and of a hybrid heuristic are about 0.01% and that of parallel scatter search even smaller.