Machine Learning based Indoor Localization Techniques for Wireless Sensor Networks

Abstract

Indoor localization is the vital technique for wireless sensor networks (WSNs) and is used in various applications where location of the sensor node is also essential. Current localization system such as global positioning system (GPS) is costly and lacks accuracy due to non-line of sight conditions inside the building. Therefore, widespread research work has been carried out on indoor localization techniques. In general, the indoor localization techniques are categorized into: a) range-based and b) range-free localization. In the last few years, research has been shifted towards cost effective accurate indoor localization using machine learning. In this paper, a new taxonomy called machine learning based indoor localization (MLBIL) techniques is recommended for the same. Moreover, an extensive survey of MLBIL techniques is also presented. Several such techniques with promising localization results are analyzed in this paper. Most of the key machine learning algorithms are utilized for MLBIL. Based on the feature(s) used, MLBIL techniques can be further categorized into three: a) Received signal strength (RSS) learning b) Non-RSS learning and c) Multiple features’ learning. Key localization challenges are highlighted in this paper. Issues of anchor nodes’ deployment, overall cost of the system, boundary node problem, and inaccurate measurements are also discussed. Support vector machine (SVM) algorithm and received signal strength indicator (RSSI) feature are commonly used for localization. It is because of easy availability of RSSI and suitable features of SVM e.g. kernel method and similarity function. It is recommended to use multiple features’ learning under MLBIL for localization in complex scenarios. It is observed that use of appropriate filter along with machine learning algorithm can increase the localization accuracy.