The Ant Colony System (ACS) is a well-known bio-inspired optimization algorithm which has been successfully applied to several NP-hard optimization problems, including transportation network optimization. This paper introduces a method to improve the computational time required by the algorithm in finding high quality solutions. The purpose of the method is to predict the best termination iteration for an unseen instance by analyzing the performance of the optimization process on solved instances. A fitness landscape analysis is used to understand the behavior of the optimizer on all given instances. A comprehensive set of features is presented to characterize instances of the transportation network optimization problem. This set of features is associated to the results of the fitness landscape analysis through a machine learning-based approach, so that the behavior of the optimization algorithm may be predicted before the optimization start and the termination iteration may be set accordingly. The proposed system has been tested on a real-world transportation network optimization problem and two randomly generated problems. The proposed method has drastically reduced the computational times required by the ACS in finding high quality solutions.