Improving Indoor Positioning System using Machine Learning Techniques

The proposed paper is to research indoor human coordinate estimation by the channel state information (CSI) method and improve the accuracy of this estimation through machine learning techniques. CSI technology provides a variety of channels, which makes it possible to analyze the data in detail. In our experiment environments, we considered two patterns of transmitter and receiver position for finding good CSI detection under device-free localization. The first pattern is both devices being in the corner of the room (opposite each other), and the second pattern is the transmitter and receiver being put in the middle and corner of the room, respectively. Four types of machine learning are considered for position estimation: artificial neural networks (ANN), K-nearest neighbor (KNN), decision tree, and Naïve Bayes. First, we examined the correlation of all features in the input data and discovered that they have a low correlation with one another. Our experimental findings show that the second pattern of transmitter position has greater precision than the first pattern of transmitter position. In addition, the placement accuracy of the second pattern's ANN is 80.5%, which is the highest performance accuracy of all other methods because the ANN is suited for low correlation data applications.