Performance Evaluation of Machine Learning in Wireless Connected Robotics Swarms

Abstract

The transmission and negotiation of the robot swarms and the technical support provided by the communication technology are inseparable, and modulation recognition plays an important role in this transmission. Due to the diversity of current classifiers, choosing the correct classifier to improve the classification effect has become a key issue. To this end, this paper explores the performance of different classifiers in modulation recognition, selects six classifier families, and compares a total of 77 different single classifiers on the modulation dataset, which is implemented on the following three platforms: Weka, Python and MATLAB. The results show that the strong classifiers formed by the combination of weak classifiers is very effective, and Boosting, Bagging, and Random Forest are the three best classifier families. In addition, it was found that as the signal-to-noise ratio (SNR) increases, the overall performance of the classifier families gradually improves, but the ranking of the families performance remains consistent.