jismart2023022: Comparison of Room Occupancy Prediction Models in A Teaching Building Based on Machine Learning

Abstract

The indoor environment and human behavior are critical factors impacting energy consumption. Taking a university building in the east of China as an example, the current study aimed to develop prediction models for room occupancy based on the indoor environment. An integrated Internet of Things (IoTs) device was employed to collect 2013 sets of data, including room occupancy and corresponding indoor environmental parameters in classrooms. A correlation analysis was used to extract key features, and prediction models for indoor occupancy behavior in a typical building were developed using the decision tree, k-nearest neighbor (KNN), random forest, and CatBoost models. Various evaluation indicators, including F1 scores, accuracy, precision, recall, and AUC values were employed to compare the performance of each model. The results revealed that the CatBoost model exhibited superior predictive capability, while the KNN model performed the least effectively. The current study provides a reference point to detect human behavior.