Toward Fast and Accurate Sensor Data Prediction Using iDC-MLP Algorithm for Industrial IoT

This paper presents a time-efficient-based approach for data prediction algorithms by utilizing lightweight Deep Learning (DL) techniques in the Industrial Internet of Things (IIoT) networks. Recent works mainly focus on data prediction accuracy without concerning time efficiency. However, real-time response is required in time-critical scenarios to predict missing data and avoid unwanted issues. Hence, a fast prediction model is mandatory to satisfy that condition. Not only for data prediction but an accurate DL model can also be used to recover missing sensor data and extend device lifetime by reducing data retransmission. An efficient DL model, called the improved Deep Concatenation Multi-Layer Perceptron (iDC-MLP), was exploited to carry out fast and reliable data prediction and recovery. The proposed iDC-MLP model was evaluated using various performance metrics under 10-fold cross-validation settings to demonstrate its robustness. Simulation work shows that the proposed iDC-MLP performs better than the existing solution with an average 18.66% error reduction in the dynamic environment. In addition, based on the experimental work conducted, the proposed model is 84.54% faster in producing single data prediction than other models. Finally, by implementing the iDC-MLP model, a total of 2.21% CPU utilization and 58.93% of network delay is successfully reduced.