Indoor Positioning via Gradient Boosting Enhanced with Feature Augmentation using Deep Learning

Abstract

With the emerge of the Internet of Things (IoT), localization within indoor environments has become inevitable and has attracted a great deal of attention in recent years. Several efforts have been made to cope with the challenges of accurate positioning systems in the presence of signal interference. In this paper, we propose a novel deep learning approach through Gradient Boosting Enhanced with Step-Wise Feature Augmentation using Artificial Neural Network (AugBoost-ANN) for indoor localization applications as it trains over labeled data. For this purpose, we propose an IoT architecture using a star network topology to collect the Received Signal Strength Indicator (RSSI) of Bluetooth Low Energy (BLE) modules by means of a Raspberry Pi as an Access Point (AP) in an indoor environment. The dataset for the experiments is gathered in the real world in different periods to match the real environments. Next, we address the challenges of the AugBoost-ANN training which augments features in each iteration of making a decision tree using a deep neural network and the transfer learning technique. Experimental results show more than 8% improvement in terms of accuracy in comparison with the existing gradient boosting and deep learning methods recently proposed in the literature, and our proposed model acquires a mean location accuracy of 0.77 m.

SANGRIA: Stacked Autoencoder Neural Networks With Gradient Boosting for Indoor Localization

“…The work in [7] combines gaussian process classifiers (GPC) with an augmentation strategy to deal with noisy RSS data, but for a simpler localization problem with lower resolution. Several recent ML techniques have used gradient-boosted trees (GBT) for indoor localization [8]- [10]. The GBT algorithm in [8] is applied on BLE RSS data.…”

“…Several recent ML techniques have used gradient-boosted trees (GBT) for indoor localization [8]- [10]. The GBT algorithm in [8] is applied on BLE RSS data. However, Wi-Fi RSS is more effective for localization compared to BLE RSS due to its wide availability, high accuracy, low cost, and longer range compared to BLE [9].…”

“…A few prior works have used augmentation techniques for indoor localization. In [12], generative adversarial networks (GANs) are used as an augmentation module, while [7] employs stacked autoencoders (SAEs), and [8] employs a simple neural network.…”