An Efficient Missing Data Prediction Method Based on Kronecker Compressive Sensing in Multivariable Time Series

The existence of missing data severely affects the establishment of correct data mining model from the raw data. Unfortunately, most of the existing missing data prediction approaches are inefficient to predict missing data from multivariable time series due to the low accuracy and poor stability property. To address this issue, we propose an efficient method using the novel Kronecker compressive sensing theory. First, we exploit the spatial and temporal properties of the multivariable time series to construct the sparse representation basis and design the measurement matrix according to the location of missing data. Accordingly, the missing data prediction problem is modeled as a sparse vector recovery problem. Then, we verify the validity of the model from two aspects: whether the sparse representation vector is sufficiently sparse and the sensing matrix satisfies the restricted isometry property of compressive sensing. Finally, we investigate the sparse recovery algorithms to find the best suited one in our application scenario. Simulation results indicate that the proposed method is highly efficient in predicting the missing data of multivariable time series.