SGD-Based Wiener Polynomial Approximation for Missing Data Recovery in Air Pollution Monitoring Dataset

This paper describes the developed SGD-based Wiener polynomial approximation method for the missing data recovery of air pollution monitoring tasks. The main steps of algorithmic implementation of the method have been described and the necessity of a combination of both of these tools is substantiated. The basic parameters of the method (the degree of the polynomial, the loss function of the SGD algorithm) for design an optimal variant of it are experimentally investigated. One out of four studied loss functions was chosen for the practical implementation of the method for the design of the future applied air pollution monitoring system. It is founded that high degrees of the Wiener polynomial significantly increase the training time with a slight increase in accuracy. That’s why a second-degree polynomial was chosen. The simulation of the method showed high as accuracy (based on MAPE, RMSE, MAE) and low computation time. Comparison of the developed method’s results with the existing regression analysis methods (Adaptive Boosting, GRNN, SVR with different kernels) confirmed the high efficiency of its work. The proposed combination of the method allows obtaining an effective result from the point of view of accuracy-speed for the large volumes of data processing. The developed method will be useful when solving different tasks, for example, for a smart home or a smart city, medicine, economics, etc. That is, for those tasks where the problem of missing data does not allow conducting further effective intellectual analysis.