Abstract: An fully connected neural network(FCNN) approach is utilized in this study to construct a grading evaluation model for concert hall acoustics, which can evaluate their quality accurately and efficiently with acquired acoustic objective parameters. By inputting three categories and thirteen types of acoustic objective parameters of concert halls, along with the corresponding quality grades outputs, a FCNN model was trained resulting in a highly accurate grading evaluation model for concert hall acoustics. The trained concert hall acoustic quality grading evaluation model is demonstrated exceptional precision in evaluating the grading with a coefficient of determination (R²) of 1. Compared to traditional grading evaluation methods, the computation time of the FCNN-based grading evaluation method is about one-tenth of the former. Additionally, with analysis of weight matrices between the input layer and the hidden layer, and decision tree algorithms based on machine learning, the weighted optimization of 13 acoustic objective parameters is conducted. Ultimately, significant acoustic parameters that impact the acoustic quality grades of concert halls were identified and ranked. In the ranked results of acoustic quality evaluation of concert halls, category of temporal acoustic parameters exhibit noticeably higher weightage compared to other categories. The research results show that FCNN method can minimize the influence of subjectivity of traditional grading methods. It can be applied in optimizing concert hall design and enhancing the audience experience.