Public audits in universities have exposed fraudulent practices in scientific research expenditure, posing a significant obstacle to progress. To address this issue, it is imperative to conduct risk assessment and analysis, thereby improving fund control and advocating for standardized research cost management. Such measures are crucial in alleviating the burden on institutions and researchers, fostering a more effective and efficient scientific research environment. In this study, an analysis of the factors affecting scientific research performance yielded three key elements: external environment, individual researchers, and information platform. After applying the nonlinear mapping ability and adaptability of back-propagation artificial neural network (BP-ANN) reverse neural network and obtaining simulation results for the generalization function of discrete information, we established a model for research performance evaluation. Subsequently, research cases were selected to conduct training and fitting experiments, ultimately scoring research performance through a comprehensive evaluation process. The experiment showed that the prediction results obtained using the BP-ANN algorithm, following learning from preprocessed samples, exhibit high accuracy. Moreover, these results can be updated and adapted with new sample inputs, highlighting the strong feasibility of this method.