Deep neural networks for variable selection of higher-order nonparametric spatial autoregressive model

Deep neural network technology has been receiving increasing attention and being applied in numerous fields due to its strong prediction performance and generalization ability. This paper examines the variable selection problem on the higher-order nonparametric spatial autoregressive model with nonparametric endogenous effects, using an efficient deep learning method. To achieve simultaneous parameter learning and variable selection, the deep learning method applies the concept of the Lasso penalty. In different data distribution scenarios, this method demonstrates strong competitiveness when compared to some variable selection methods. Simulation experiments and real data analysis reflect the effectiveness of this method.