An Efficient Localization for Indoor Environment using Classification Algorithms

The positioning accuracy of various kinds of devices is often calculated by using Wi-Fi fingerprints as a data source of Received Signal Strength (RSS). Through different matching algorithms, these RSS values are related to corresponding positioning coordinates for positioning estimation. In this paper, we compare two well known Machine Learning (ML) structures for calculating positioning accuracy: Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM) and a Decision Tree (DT). Both matching algorithms run through a series of datasets in two parts. In the first part, the datasets contain raw, non-normalized RSS values. In the second part, the RSS values are all normalized to remove noise in the datasets. After extensive testing and simulations, the tests showed that as long as RSS values are first normalized, a Decision Tree offers better average positioning accuracy.