Global optimal model selection for high-dimensional survival analysis

With the popularity of high-dimensional data, model selection is of great importance in recent survival analysis. In a model selection context, an important research question is how to define the best model. To answer this, various model selection criteria have been proposed for defining the best model. The existing methods commonly use the L-0-norm penalization in order to measure the model complexity based on the number of parameters. However, due to the nonconvexity of the L-0-penalty, finding the best model via global optimization has been a challenging research subject in statistics and machine learning. In this paper, we propose a global optimization algorithm using a modification of the simulated annealing, which is a probabilistic search algorithm for the global optimum in statistical mechanics. The performance of the proposed method is examined via simulation study and real data analysis.