Abstract: Aiming at the supply-demand imbalance caused by the mismatch of rural domestic waste management modes, this study constructs a K-means-BP neural network model under the perspective of township differentiation, and achieves scientific preference by predicting the fitness of management modes. The results show that: 1) Based on K-means clustering, the 122 townships in Shaanxi Province are classified into three categories: urban suburban high-density type (24.59%), urban-rural transition medium-density type (45.08%), and remote mountain low-density type (30.33%), and summarised to collect three typical modes in Shaanxi Province: the government-enterprise cooperation mode, the county government governance mode, and the township selfgovernance mode; 2) Through rooting theory, 10 indicators such as population density, resource recycling rate, and degree of environmental impact are extracted as indicators for model selection to reveal the mechanism of interaction effects on model fitness; 3) The model prediction results show that the most suitable model for Type I townships is the government-enterprise cooperation governance model, with a fitness of 92.77%; The most suitable model for Type II townships is the county government governance model, with a fitness of 85.34%. Accordingly, differentiated governance suggestions are made to provide a decision-making basis for improving rural waste governance.