Improving the Efficiency of Predictions Based on the Analysis of Monitoring Data in the Maintenance of Smart Buildings

Data processing of monitoring systems of various processes at the stage of construction and maintenance of buildings requires the development of special tools that belong to the field of artificial intelligence. This article explores how different time series data preprocessing approaches, in particular detrending and seasonality removal, affect the accuracy and performance of computational intelligence models. Three variants of data preprocessing methods are considered: detrending, deseasonalization and their combination. From the experiments conducted on four data sets, three main conclusions are made: 1) removing the trend and seasonality separately does not improve overall performance, 2) removing the trend can greatly impair the accuracy of the model, and 3) simultaneous use of detrending and deseasonalization has a positive effect in improving the accuracy of models.