Machine Learning for Indoor Localization using CSI Technique

In this research, we propose a low-cost indoor localization technique using the CSI. By using CSI signal as input data, different locations and human activities are classified effectively using machine learning models. According to a low-cost constraint where a single transmitter (Tx) and receiver (Rx) are utilized, we investigate the appropriate transmitter location for achieving accurate position estimation in an indoor environment under two patterns of room environment. Moreover, an appropriate machine learning model for position estimation is examined where decision tree, KNN, Naïve bayes, and ANN are considered. The experimental results demonstrate that the highest accuracy can be achieved using the ANN technique. The shortest distance between a Tx and Rx provides the highest precision for determining the human location in an empty room, whereas placing the Tx and Rx in opposing positions is advantageous for a section with items positioned against the wall. When determining an average accuracy, placing the Tx and Rx in opposing positions shows the highest accuracy among others.