Data Imputation on Lost IoT Data

The amount of data resulting from IoT-based industrial applications is growing rapidly nowadays. However, due to failures and communication breakdowns in IoT devices, noise, uncertainty and incompleteness may occur in the collected data. This issue has become a critical issue for data generation, quality, processing and analysis. Incomplete or inaccurate heat meter data in central heat cost sharing system, which has an important place in energy efficiency, is one of the biggest problems in making a fair share. In this study, based on the heat meter data retrieved from Pro Tek Energy Services company serving in Sivas, 8 different machine learning algorithms including Linear Regression, Polynomial Regression, K-Nearest Neighborhood, Support Vector Machines, Random Forest, Decision Tree, Adaboost and Multilayer Perceptron were used to predict the average daily outdoor temperature and independent unit of energy consumption. As a result of the experimental studies of these algorithms were evaluated based on R2, RMSE and MAPE metrics. It is observed that the performance of different algorithms stands out in different data sets. © 2023 IEEE.