Evaluating Link Lifetime Prediction to Support Computational Offloading Decision in VANETs.

Area of Science:

• Intelligent Transportation Systems (ITS)
• Vehicular Ad Hoc Networks (VANETs)
• Machine Learning (ML) in vehicular communication

Background:

• VANETs enable intelligent applications for urban mobility, enhancing traffic management and safety.
• Computational offloading is crucial for VANET applications due to limited onboard processing power.
• Dynamic vehicle mobility results in short communication link lifetimes, challenging offloading decisions.

Purpose of the Study:

• To investigate machine learning algorithms for predicting link lifetime in VANETs.
• To improve the accuracy of communication time estimation for computational offloading.
• To evaluate the effectiveness of ML-based link lifespan prediction in road and urban scenarios.

Main Methods:

• Investigated five machine learning algorithms for link lifetime prediction.
• Developed procedures for creating training datasets for ML models.
• Trained and evaluated Support Vector Regression (SVR) and XGBoost algorithms.

Main Results:

• ML-based regression approaches significantly decreased prediction error rates compared to conventional methods.
• SVR effectively reduced task loss and improved recovery rates in computational offloading.
• XGBoost demonstrated superior performance, reducing task recovery or drop rates by 70-80%.

Conclusions:

• Accurate link lifetime prediction using ML enhances the efficiency of computational offloading in VANETs.
• SVR and XGBoost show significant potential for improving VANET application performance.
• This research provides a foundation for more efficient data estimation in diverse vehicular settings.