Graph regularized local self-representation for missing value imputation with applications to on-road traffic sensor data

摘要: Abstract Recovering missing values (MVs) from incomplete data is an important problem for many real-world applications. Previous research efforts toward solving MVs primarily exploit the global and/or local structure of data. In this work, we propose a novel imputation method by combing sample self-representation strategy and underlying linear in uniformed framework. Specifically, proposed consists following steps. First, existing applied to obtain first-round estimation MVs. Then, graph, characterizing proximity data, constructed based on imputed Next, model coined as graph regularized (GRLSR) integrating two crucial elements: regularization. The former assumes each can be well represented (reconstructed) linearly combining neighboring samples while latter further requires should not deviate too much other after reconstruction. By doing so, more accurately restored due joint We also develop effective alternating optimization algorithm solve GRLSR model, thereby achieving final imputation. convergence computational complexity analysis our are presented. To evaluate method, extensive experiments conducted both traffic flow dataset UCI benchmark datasets. results demonstrate effectiveness compared with set widely-used competing methods.