Indoor Positioning System using WKNN and LSTM Combined via Ensemble Learning

Abstract

Indoor positioning system (IPS) has become a high demand research field to be developed and has made considerable progress in recent years. Wi-Fi fingerprinting is the most promising technique that produces an acceptable result. However, despite the large amount of research that has been done using Wi-Fi fingerprinting, only a few Wi-Fi based IPS in the market can be said to be successful. Doing the research in a controlled environment and ignore the temporal signal changes may be the cause of such scenario. A long-term dataset was built to overcome this issue, yet the distance error of the state of the art was 2.48m. Therefore, we aim to reduce the distance error by combining two positioning algorithms which are Weighted k-Nearest Neighbor (WKNN) and Long-Short Term Memory (LSTM) using ensemble learning. The result shows that our ensemble method can reduce the localization error to 1.89m and improve the performance of the IPS by 23.7% when compared to the state of the art.