A MINERAÇÃO DE DADOS PARA SELEÇÃO DE HEURÍSTICAS NO PROBLEMA DE EMPACOTAMENTO BIDIMENSIONAL RETANGULAR

Abstract

The rectangular two-dimensional strip packing problem is to position a set of small rectangles within a fixed width and virtually infinite length range, while minimizing the length required to position all rectangles. The overall objective of this project is to fit algorithm classification models capable of accurately selecting the best improvement heuristic option according to the characteristics of each instance of the rectangular twodimensional strip packing problem. The research methodology is based on the use of supervised data mining techniques for the adjustment of algorithm classification models. The focus is on the selection of heuristics that have the greatest potential to find quality solutions to the problem, using only as information the characteristics of the instances used. After conducting the research, it was possible to observe that the supervised data mining techniques support vector machine with polynomial kernel is the most efficient in the search for the best improvement heuristic option for the problem context. Still, it was noted that the characteristics of the instances are able to send important information to solve the problem.