Performance Comparison of Indoor Fingerprinting Techniques Based on Artificial Neural Network

Abstract

An indoor fingerprinting localization algorithm based on a single Artificial Neural Network (ANN) model may be subject to the Received Signal Strength Indicator (RSSI) fluctuation than multiple neural networks based fingerprinting algorithm. To date, there has not been an adequate analytical study that has investigated the performance comparison of both models. In this work, a multiple neural network fingerprinting localization model is proposed which predicts the estimated position using the result of the combination of several single neural networks. Additionally, a K-nearer coarse localizer is used to perform the position estimation task. This model has been evaluated by comparison with the existing K-Nearer Neighbor (KNN) based method and single neural network based fingerprinting localization method on a corridor and Office area and on publicly available RSSI database. The results of the proposed model are more accurate than those with KNN and single neural network model. We believe that the multiple neural networks model will be more robust for positioning algorithms using several types of features data at the same time.