A Robust UWSN Handover Prediction System Using Ensemble Learning.

Area of Science:

• Marine technology
• Wireless communication networks
• Data science

Background:

• Underwater wireless sensor networks (UWSNs) are increasingly vital for marine data collection and monitoring.
• Handover prediction in UWSNs presents unique challenges due to distinct sensor node mobility compared to terrestrial networks.
• Existing research has limited focus on handover prediction within UWSNs, particularly concerning ensemble learning applications.

Purpose of the Study:

• To address the critical challenge of handover prediction in underwater wireless sensor networks (UWSNs).
• To investigate the efficacy of machine learning algorithms, including ensemble learning, for predicting handover events in UWSNs.
• To develop and evaluate advanced handover prediction models utilizing realistic marine environmental data.

Main Methods:

• Simulated UWSN environments incorporating sensor node mobility based on real marine data (water current speed and direction).
• Generation of a comprehensive dataset of handover events from the UWSN simulation.
• Application and evaluation of four machine learning algorithms: gradient boosting, decision tree (DT), Gaussian naive Bayes (GNB), and K-nearest neighbor (KNN) for handover prediction.

Main Results:

• Achieved prediction accuracy rates exceeding 95% for handover events in UWSNs.
• Demonstrated a significant improvement over the state-of-the-art method, which reported 56% accuracy.
• Ensemble learning and decision tree models showed comparable high performance, outperforming GNB and KNN.

Conclusions:

• The proposed machine learning models offer a highly accurate solution for handover prediction in UWSNs.
• Simulation using real marine data effectively captures UWSN mobility dynamics for robust model training.
• The study highlights the potential of ensemble learning and decision trees for enhancing UWSN reliability and performance.