Recovery of Incomplete IoT Sensed Data using High-Performance Extended-Input Neural-Like Structure

Abstract

The task of the recovery of incomplete IoT sensed data is considered. A high-performance extended-input neural network, which is constructed using a non-iterative neural-like structure and a Wiener polynomial for its solution, has been applied. Modelling of its work on the data collected by the IoT device was conducted. The research of hyperparameters selection of the proposed technique for missed data recovery was carried out. The number of neurons in the ANN’s hidden layer and the polynomial degree for the effective method operation were selected by experimental way. It is established the significant reduction of the training time as well as the accuracy increase by reducing the ANN’s hidden neurons number. A comparison of the proposed technique effectiveness with the state-of-the-art has been carried out. The highest accuracy of its work is established in comparison with existing methods. The training time of the proposed technique was determined and compared with existing methods.