A machine learning approach to synchronization of automata

We present a novel method to predict the length of the shortest synchronizing words of a finite automaton by applying the machine learning approach. We introduce several so-called automata features which depict the structure of an automaton, and use them with machine learning algorithms. The article discusses effectiveness of the machine learning approach in predicting the length of the shortest synchronizing words. We also examine the impact of particular features on this length, which may be helpful in methods of constructing automata as models of real systems, algorithms finding synchronizing words, and further theoretical research on synchronizing automata and the Černý conjecture.