Performance Analysis of Machine Learning Algorithms on Self-Localization Systems

Abstract

The paper evaluates the performance of various machine learning techniques for localization systems. A case of outdoor localization based on multiple Received Signal Strength Indication (RSSI) values is considered and localization accuracy is determined for various SNR levels. Machine learning algorithms are deployed to make the system terrain aware by adapting RSSI values with the change in environment. Finally, this paper presents a performance comparison of different classifiers available in machine learning toolkit WEKA in selecting the most suitable radio frequency propagation models from a set of models. Our results show that terrain identification can be achieved using random forests and random committee classifiers within an error bound of 10 percent.