Abstract
The Indoor Positioning System (IPS) is used to locate devices and people in smart environments. In recent years, position determination methods have evolved from simple Received Signal Strength Indicator (RSSI) measurements to more advanced approaches such as Channel State Information (CSI), Round Trip Time (RTT), and Angle of Arrival (AoA), increasingly combined with Machine Learning (ML). This article presents a systematic review of the literature on ML-based IPS using IEEE 802.11 Wireless LAN (Wi-Fi) and Bluetooth Low Energy (BLE), including studies published between 2020 and 2024 under the Preferred Reporting Items for Systematic Reviews and Meta-Analyse (PRISMA) methodology. This study examines the techniques used to collect measurements and the ML models used, and discusses the growing use of Deep Learning (DL) approaches. This review identifies some challenges that remain for the implementation of these systems, such as environmental variability, device heterogeneity, and the need for calibration. Future research should expand ML applications to RTT and AoA, explore hybrid multimetric systems, and design lightweight, adaptive DL models. Advances in wireless standards and emerging technologies are also expected to further enhance accuracy and scalability in next-generation IPS.