A Parameter-Optimized Random Forest Method for CSI based Passive Indoor Occupancy Detection

This paper examines the question of the passive indoor occupancy detection based on channel state information (CSI) data. At the outset, we propose an antenna averaging and dividing (AAD) method for boosting the quality of CSI data obtained from commercial Wi-Fi devices. Then, utilizing the preprocessed data as the input, the prediction classification is carried out using a Random Forest (RF) model, employing the Bitterling Fish Optimization (BFO) method to optimize the RF parameters. Furthermore, a window-based statistical method is proposed to further enhance the prediction accuracy. Finally, we construct a CSI acquisition platform that utilizes a desktop computer equipped with an Intel 5300 NIC and a TP-LINK router; experimental results assess the dependability of the proposed algorithm using the dataset of Track 1 of Huawei Cloud’s First Wi-Fi Perception Contest.