Runtime Prediction of Optimizers Using Improved Support Vector Machine

The aim of this paper is to propose a machine learning approach to build a model for predicting the runtime of optimization algorithms as a function of problem-specific instance features. That is, our method consists of building a support vector machine (SVM) model incorporating feature selection to predict the runtime of each configuration on each instance in order to select the adapted setting depending on the instance. Such approach is useful for both algorithm configuration and algorithm selection. These problems are attracting much attention and they enable to benefit from the increasing volume of data for better decision making. The experiment consists of predicting algorithm performance for a well known optimization problem using the regression form of SVM.