Parameter estimation of INDSCAL models using simulated annealing

Parameter estimation of INDSCAL by CANDECOMP method has three disadvantages: finding the global minimum is not guaranteed, it can find negative weights and a diagonal matrix D such that X = YD may not exist (this is known as the symmetry problem). To overcome the last difficulties, Ten Berge et al. proposed an algorithm called SYMPRES. We propose in this article a new method based on the simulated annealing technique, called ssINDS, that solves the symmetry problem and the non-negativeness of the weights. Moreover the corresponding algorithm converges to the global optimum with probability one. From our computational experiment we obtained results which are similarity to SYMPRES and CANDECOMP methods.