IIOT Prediction Accuracy Improved By T-Distributed Stochastic Neighborem Bedding and Jaguar Algorithm With Non-Linear Analysis

Abstract

In the fast changing terrain of Industrial Internet of Things (IIoT), predictive accuracy is very vital to maximize operations, reduce downtime, and increase process optimization. Typical prediction models, on high-dimensional data, however, may fail, leading to less than perfect accuracy and performance. Together with non-linear analysis, this paper examines the mix of T-Distributed Stochastic Neighbour Embedding (t-SNE) and the Jaguar approach to increase prediction accuracy in IIoT systems so addressing this problem. Managing complex, non-linear interactions present in IIoT data presents the main difficulty. While conventional methods may fail in the complicated patterns prevalent to IIoT systems, they operate well in linear conditions. This paper proposes a new approach whereby t-SNE is employed for dimensionality reduction, therefore retaining the local structure of data and so minimizing the curse of dimensionality. Then the prediction model is refined using a nature-inspired optimization technique called the Jaguar algorithm, hence improving its flexibility in the face of data non-linearities. The proposed method was evaluated on a comprehensive IIoT dataset and shown to yield appreciable expected accuracy gain. The model especially outperformed traditional methods such Genetic Algorithms (GA) and Principal Component Analysis (PCA), which generated accuracies of 87.3% and 89.1%, respectively, with an accuracy of 94.8%. Moreover proving its ability to handle complex IIoT data, the proposed approach showed a 10.3% increase in precision and a 12.5% increase in recall.