Estimating occupancy in residential building by machine learning

Abstract

The development of indoor environments for humans consumes a lot of energy. As such, an accurate occupancy prediction model is crucial for increasing the effectiveness of energy simulation and occupant-centric building management. It also contributes to the completion of the life cycle phase of buildings, which includes design, operation, and retrofitting. This project focuses on developing a machine learning model to estimate the number of persons in a room ranging from 0 to 3. To handle this time series dataset, 4 machine learning techniques for type categorization are used. Feature selection is a crucial pre-processing step that is typically used to identify the most significant input characteristics in order to decrease classifier complexity and computing requirements. The recently developed swarm intelligence algorithm called Grey Wolf Optimizer (GWO) was employed in this study. The GWO assisted in identifying the six most important and influential elements for the study's findings. As giving the most optimal results, Random Forest with GWO was chosen as the proposed algorithms. Following that, grid search is utilized to enhance the results even further. The acquired findings are quite strong and encouraging, with an accuracy of more than 99.7% and an F1 score of more than 98.9%. This might be interpreted as a positive indication to try further iterations of these synthesis methods in the future in an effort to enhance the accuracy of the prediction of the room's occupancy.