A Non-Iterative Neural-Like Framework for Missing Data Imputation

Abstract The paper presents developed by the authors a non-iterative, neural-like framework for filling missing data. This framework applies the high-speed linear neural-like structure using the successive geometrical transformations model. The use of this computing intelligence tool decreases the duration of the training process, compared to known algorithms of solving the defined task. Besides, the developed framework provides a high degree of accuracy in solving the problem of filling missing data. The simulation of the method had conducted on the monitoring data of environmental pollution, where the task of fill missing data is very relevant. The high accuracy of the developed supervised learning-based framework based on the definition of RMSE, MAPE, MAE errors in recovering missing data was confirmed. Also, experiments with models were conducted comparing the duration of the training procedures of all investigated methods. The high efficiency of the developed framework was confirmed. Modern Data-driven methods have most often faced with the problem of filling out missing data in various fields. Thus developed framework can be used in medicine, information security, economics, service sciences, materials science.