Additive manufacturing technology has greatly improved the design flexibility and accelerated the optimization verification of structure that cannot be easily and economically produced by traditional subtractive manufacturing processes. However, common defects such as surface roughness and porosity, affect the quality and reliability of the components, hindering their wide application. In this study, an integrated framework incorporating high-fidelity powder-scale mechanistic model and physics-informed machine learning is developed to predict the built quality of aluminum and to determine the hierarchy importance of mechanistic variables for different printing qualities in a multi-classification problem in the processing space. The influence of different processing parameters on the built quality is explored by the mechanistic model. A decision tree is constructed and the quality prediction index (QPI) connecting five variables and the printing quality is established. The hierarchy importance of the mechanistic variables is determined by the QPI and three machine learning inductions. The most important factor for balling, good printing quality, keyhole and lack of fusion defects are Fo, TP, fr and Tp, respectively. As the mechanistic variable values are the comprehensive results of multiple processing parameters, this hierarchy ranking not only deepens the scientific understanding of different phenomenology, but also provides new insights and strategies for the process optimization.