A PSO-optimized GM(1,N) power model with hybrid feature selection for warship R&D cost prediction

This paper proposes an integrated forecasting method combining FGRA-LASSO hybrid feature selection with a Particle Swarm Optimized GM(1,N) power model to address small sample sizes, high dimensionality, and strong non-linearity in warship R&D cost prediction. The FGRA-LASSO method identifies 12 key cost drivers from subjective and objective hybrid features. Particle Swarm Optimization globally optimizes the background value weight and power exponent, constructing an adaptive grey prediction model. Experimental results demonstrate excellent predictive performance with a test set MAPE of 2.77%, significantly outperforming traditional grey models and various machine learning methods in both accuracy and robustness. This framework provides an effective theoretical tool for accurate warship R&D cost prediction.