Occupancy Detection Using Wi-Fi in Indoor Environments

Sammanfattning

Using Wi-Fi to detect occupancy could benefit smart buildings in areas such as energy management or security by utilizing already installed infrastructure. In this study, Channel State Information (CSI) data was processed and used to train machine learning classifiers, specifically Decision Trees and Support Vector Machines (SVM) with a linear kernel, to detect human occupancy in a dynamic office environment. The impact of data normalization, feature count, and varying sliding-window sizes during feature extraction on model performance was also analyzed. The decision tree classifier achieved up to 98% accuracy, while the SVM achieved up to 68%. Data normalization and increasing the number of features beyond a necessary subset were found to reduce model performance and increase training time. In contrast, larger window sizes during feature extraction consistently improved the accuracy and efficiency of the decision tree models.