Backpropagation Neural Network for Science and Technology Finance Evaluation Model for Citylevel Stf Assessment

In recent years, evaluating the development level of city Science and Technology Finance (STF) has become vital in formulating targeted innovation strategies and enhancing urban competitiveness. Traditional statistical approaches often lack the capacity to model the complex non-linear interactions between multi-dimensional indicators. To address this, this research proposes a Backpropagation Neural Network for SFT Evaluation Model (BP-STFNet) model specifically designed for city-level STF assessment. Initially, a comprehensive multi-year panel dataset was constructed by integrating indicators from three domains: science and technology, financial development, and economic support (Gross Domestic Product (GDP) per capita, infrastructure index). Then, these data undergo MinMax normalization to eliminate scale disparities, followed by redundancy reduction using correlation analysis and factor analysis to ensure computational efficiency. Feature selection and optimization combine the Pearson correlation for mutual information analysis and Principal Component Analysis (PCA) to extract the most relevant indicators while preserving interpretability. The optimized indicator set is input into a BP neural network with sigmoid activation in hidden layers and Softmax for multiclass output. The experimental results demonstrated that BP-STFNet outperformed the existing simple BP neural network in terms of a classification accuracy of 87.94 %.