Robust forecasting in spatial autoregressive model with total variation regularization

In recent decades, feature selection has attracted great attention in data science due to curse of dimensionality, which arises in larger, complex, and heterogeneous data. However, existing researches of feature grouping on spatial autoregressive model are rare. To address this challenge, this paper investigates robust spatial autoregressive model with feature grouping and robust forecasting achieved automatically. The proposed novel methodology borrows strength from check loss function in quantile regression and total variation regularization. A simple-to-implement algorithm following double-level alternative method of multipliers design is derived computationally. The empirical studies demonstrate the effectiveness of the proposed methods via comparing with other competing forecasting techniques.