Estimation of real-valued functions from a finite set of samples is a central problem in applied statistics. Many different approaches to deal with this problem were proposed as the least-squares method by Gauss, the least-modulus method by Laplace, and more recently the usage of neural networks, support vector machines, mixture of local expert models, amongst others. We addressed the issues mixture of local expert models (MLEM) and clustering optimization, which congregates exploratory analysis, data mining and mathematical modeling in the same technique, used, for example, in the development of predictive models. The basic idea of MLEM is clustering the points from the entry data set, and then different modeling techniques are applied in order to select the best model for each cluster. We proposed a clustering optimization procedure as a way to improve the performance on both the fitting of the models and their usage in forecasting.
mixture of experts; clustering; optimization
