Characteristic time scale as optimal input in Machine Learning algorithms: Homogeneous autoignition