A1257
Title: Multidimensional scaling by particle swarm optimization
Authors: Javier Trejos - University of Costa Rica (Costa Rica) [presenting]
Victor Bazan - University of Costa Rica (Costa Rica)
Abstract: Multidimensional Scaling (MDS) dimension reduction is performed via the metaheuristic approach of particle swarm optimization (PSO), by minimizing the Stress loss function. Particles in a low-dimensional space identify solutions of the MDS problem, and the approximation of the optimum is tuned by weighing the terms of the iterative velocity equation. In the PSO implementation for MDS, the velocity equation considers terms of inertia, past experience, imitation to the leader, and neighbors influence, with uniform random coefficients for each term. The method performs well compared to known methods such as Smacof or implementations of other metaheuristics (simulated annealing and genetic algorithms). Results are illustrated on several datasets, including a real dataset of chemical organization of products in a warehouse. The implementation was made in R.
