Integrated Online Prediction Model for IoT Data

Data prediction research is very important for applications based on the Internet of things (IoT). However, the large-scale, high-dimension, and the inevitable missing values seriously hinder accurate prediction result. To meet this challenge, an integrated online prediction model is proposed, which can solve future values prediction and missing values imputation simultaneously in the low-dimensional space. Specifically, we embed the classic prediction models into the objective function of matrix factorization (MF) to obtain low-dimensional embedding parameters, which are then used to construct the prediction model. Furthermore, a rolling prediction mechanism is proposed to update the prediction model when new sampled data comes, based on the online MF with time-varying parameters. Simulation results on four UCI public data sets show the superior performance of the proposed method.