View Abstract Sign in for premium content
The aim of this paper is to apply semi-supervised machine learning (ML) approach for modelling a classifier that can deftly classify the signals obtained from the NavIC GSO satellites using the code minus carrier observables and elevation angle parameter. As supervised ML requires labels for training the models, the labels generated from an existing work by the author to classify multipath, LOS and NLOS in the NavIC satellites using unsupervised learning approach was used to train Support Vector Machine (SVM), Decision Tree (DT) and Random Forest (RF) algorithms using NavIC L5 data collected from Dehradun, India. This work provided insight to identify the best performing model out of the three selected supervised ML models. Based on the analysis through the metrics testing accuracy, confusion matrix and cross-validation, it was observed that out of the three supervised learning algorithms attempted, the Support Vector Machine SVM algorithm outperformed DT and RF algorithms for the multipath classification of the NavIC signals. The results show that the SVM based algorithm has achieved 99.98% accuracy in terms of the multipath classification both the NavIC GSO satellites 2 and 4. Another metric cross-validation also shows that SVM is the best performing model for NavIC GSO multipath classification. This study can be useful for the development of NavIC and multi-GNSS robust and intelligent positioning algorithms & applications. Furthermore, this classification of multipath signals can also be instrumental for soil moisture estimation application, which is an important input for the flood prediction, weather forecast and climate monitoring.