<p>The issue of excessive measurement point placement in electric multiple units (EMU) framework testing has led to problems such as signal redundancy, increased testing costs, and excessive data volume. This paper proposes an optimization method for the arrangement of measurement points in EMU frameworks based on the Newton-Raphson-based optimizer-back propagation (NRBO-BP) neural network. The aim is to leverage the similarity and predictability of measurement point equivalent stress characteristics to obtain complete framework stress data with the minimum number of measurement points. This approach innovatively integrates system clustering analysis, Spearman correlation analysis, and the NRBO-BP neural network to construct a comprehensive optimization process. Firstly, the equivalent stress of measurement points is calculated using long-term testing data and fatigue damage theory. The distribution and dispersion characteristics of these points are analyzed through kernel density estimation. Secondly, measurement points are grouped using system clustering based on the principles of similar mean, standard deviation, and root mean square values for equivalent stress. Spearman correlation analysis is then used to determine the input and output measurement points for the predictive model. Finally, the NRBO-BP neural network is employed to establish a prediction model for equivalent stress of similar measurement points, thereby determining the optimal number of measurement points. The results show that the NRBO-BP neural network outperforms the BP, genetic algorithm-back propagation (GA-BP), long short-term memory (LSTM) and Transformer models across various evaluation metrics, indicating superior prediction performance and accuracy. By applying the NRBO-BP neural network, the number of measurement points can be reduced from 15 to 5. This optimization contributes to reducing experimental costs and provides technical support for the regularized in-service safety monitoring of EMU frameworks.</p>