The number of rules and parameter values in extended belief rule base(EBRB) affect the accuracy and computing efficiency of the EBRB inference model. Therefore, this paper proposes an improved EBRB inference method based on rule clustering and parameter learning, called RCPL-EBRB model. The principles of the proposed model include: The density clustering analysis is firstly used to perform the rule clustering of the EBRB, so as to identify invalid extended belief rules and improve the modeling process of the traditional EBRB. Then, the rule clusters obtained by clustering, namely sub-EBRB, are used as basic units for parameter learning and rule reasoning, so as to improve the accuracy and computing efficiency of the RCPL-EBRB model. Finally, the datasets of nonlinear function fitting and benchmark classification problems are introduced to verify the effectiveness of the proposed model and carry out parameters sensitivity analysis. Results show that the RCPL-EBRB model has higher accuracy than the existing EBRB inference model and traditional machine learning methods.