IMPROVED INDOOR LOCALIZATION WITH MACHINE LEARNING TECHNIQUES FOR IOT APPLICATIONS

With the rapid development of the internet of things (IoT) and the popularization of mobile internet applications, the location-based service (LBS) has attracted much attention due to its commercial, military, and social applications. The global positioning system (GPS) is the prominent and most widely used technology that provides localization and navigation services for outdoor location information. However, the GPS cannot be used well in indoor environments due to weak signal reception, radio multi-path effect, signal scattering, and attenuation. Therefore, localization-based systems for indoor environments have been designed using various wireless communication technologies such as Wi-Fi, ZigBee, Bluetooth, UWB, etc., depending on the context and application scenarios. Received signal strength indicator (RSSI) technology has been extensively used in indoor localization technology due to it provides accuracy, high feasibility, simplicity, and deployment practicability features. Various machine learning algorithms have been employed to find the most accurate location from the RSSI-based indoor localization. This study consists of three phases for the machine learning algorithms to solve the localization issue with different wireless technologies named the supervised regressors algorithms, supervised classifiers algorithms, and ensemble machine learning for RSSI-based indoor Localization. In addition, a weighted least squares technique and pseudo-linear solution approach are proposed for the closed-form solution. The proposed methods approximate the original system of non-linear RSSI measurement equations with a system of linear equations. Then, the experimental testbed was designed using various wireless technologies with multiple anchor nodes to explore, test, evaluate and compare in a testbed data collection. In each testbed, RSSI values were collected using IoT cloud architectures. The collected data were pre-processed investigating appropriate filters before training the algorithm. Finally, the received RSSI value is trained using several machine learning models named linear regression, polynomial regression, support vector regression, random forest regression, and decision tree regressor under various wireless technologies. It estimates the geographical coordinates of a moving target node and compares them with each supervised machine learning technique’s performance using model evaluation matrices. Consequently, the experiment?s performance outcome is expressed in terms of accuracy, root mean square errors, precision, recall, sensitivity, coefficient of determinant, and the f1-score.