Evolving feature selection for characterizing and solving the 1D and 2D bin packing problem

Abstract

This paper presents an evolutionary framework that solves the one and two dimensional bin packing problem by combining several heuristics. The idea is to apply the heuristic that is more suitable at each stage of the solving process. To select a heuristic to apply, we characterize the problem employing a number of features. It is common in many existing approaches, that the user selects a set of features to represent the problem instances. In our solution model, we start with a large set of features, and those that succeed characterizing the instances are automatically selected during the evolutionary process. After providing a list of features, the user does not have to select the features that are best suitable to characterize problem instances. Therefore our system is more knowledge independent than previous approaches. This model produces better results employing the proposed feature selection approach compared against the use of other feature selection methodology.