Missing IoT Data Prediction with Machine Learning Techniques

Abstract

Every day, the amount of data generated by industrial applications based on the Internet of Things (IoT) grows. However, data acquired as a result of failures and communication disconnections in IoT devices might be noisy, inaccurate, and incomplete. These issues have become crucial for data production, quality, processing, and analysis. The datasets used in the scope of this study were collected in real-time from the water neutralizer system of Sivas Numune Hospital, which converts medical waste into household waste. Medical liquid wastes in hospitals are exposed to chemical neutralization process by means of pH change with neutralization devices before being transferred to the sewer. In this regard, the monitoring of pH levels in the medical waste neutralization system is crucial for environmental protection. In this aspect, two datasets with varying quantities of missing data were evaluated for the prediction of the PH using the linear regression (LR), support vector machines (SVM), k-nearest neighbor (KNN), random forest (RF), and decision tree (DT) machine learning algorithms. Mean absolute error (MAE), mean squared error (MSE), and root mean square error (RMSE) performance metrics were used to evaluate machine learning algorithms. As a consequence of the analysis, it was determined that the SVM algorithm performed better performance on the two distinct datasets. The result of the evaluation indicates that machine learning algorithms are remarkably efficient at predicting missing pH data.