Visualization of Data Using Genetic Algorithm

Abstract

In order to obtain a good spline model from large measurement data, we frequently have to deal with knots as variables, which becomes a continuous, non-linear and multivariate optimization problem with many local optima. Hence, it is very difficult to obtain a global optima. In this paper, we present a method to convert the original problem into a discrete combinatorial optimization problem and solve it by a genetic algorithm. We also incorporate a comer detection algorithm to detect significant points (comer points), which are necessary to capture a pleasant looking spline fitting for 2D and 3D data. In case of too large data, a data reduction concept is also utilized to economize the computation in the algorithm design. As a curve and surface model, the parametric B-Spline model has been approximated to various 2D and 3D data. The chromosomes have been constructed by considering the candidates of the locations of knots as genes. The knots to the corresponding comer points have been kept fixed to minimize the computation cost. The best model among the candidates is searched by using Akaike’s Information Criterion (AIC). The method determines the appropriate number and location of knots automatically and simultaneously.