IoT indoor localization using design of experiment analysis and multi-output regression models

Abstract

Received signal strength indicator (RSSI) measures the power level present in a received radio signal, and it is the mainstream wireless signal measurement tool for vast indoor localization systems. In this study, the researchers addressed the lack of research in comparing popular wireless technologies using design of experiments (DOE) and lack of contrast of inherently multi-output learning regression algorithms to improve the model accuracy and set a standard algorithm for indoor localization. This would benefit IoT developers in decisioning thewireless technology and algorithm to be utilized in a specific task. The study investigated RSSI from independent factors using coded levels: interference, distances from the transmitter, and wireless technologies on publicly available IoT localization datasets. The study determined the parameters significantly affecting the RSSI through DOE and statistical analysis. Using Minitab®, the study implemented a 3k full-factorial design and analyzed the model using ANOVA, optimization and residual plots, regression equations, and model summaries. Moreover, different regression algorithms were tested: multiple linear, k-nearest neighbors, decision tree, and random forest regression methods. Performance metrics used were MSE, standard deviation, and R-sq values. Results showed that there is an optimum wireless technology for a given set of optimal conditions. Also, the DT model outperformed well for an indoor localization application.