Stacked Spatio-Temporal Graph Neural Network Model Based on Mask-Driven for Time Series Imputation

The existence of missing values in time-series data has brought great inconvenience to our lives. In this paper, we propose a Stacked Spatio-Temporal Graph Neural Network model based on Mask-Driven for time series imputation(SSTGNN-MD). The entire model is operated in the frequency domain, via converting time-domain data into frequency-domain data through wavelet transform, and uses masks to select important features of the data. The entire model mainly consists of three modules: mask module, temporal convolution module, and spatial convolution module. The three modules interact with each other to fully acquire the features of the data, in order to achieve the task of filling incomplete temporal data. The experiments on several datasets evaluate the effectiveness of proposed method for incomplete multivariate time series imputation.