This project investigates machine learning-based handover prediction in vehicular networks using hybrid deep learning models. We compare Graph Convolutional Networks combined with LSTM (GCN+LSTM) against an ensemble approach (GCN+LSTM+SVM) to predict optimal handover points in 5G mobile networks, minimizing latency and packet loss during vehicle mobility.
- Networking: 5G/LTE protocol stacks, handover mechanisms, mobility management
- Machine Learning: Graph Neural Networks, LSTM, ensemble methods, PyTorch/TensorFlow
- Simulation: Discrete event simulation, network modeling, performance analysis
- Systems Integration: Bridging ML frameworks with C++ simulators, data pipeline design
- Research Methodology: Comparative analysis, metrics definition, reproducible experiments
- GCN+LSTM: Captures temporal-spatial dependencies using graph convolutions on network topology + sequential LSTM layers for temporal patterns
- GCN+LSTM+SVM: Ensemble approach adding SVM for classification refinement, improving prediction accuracy
- Simu5G: 5G network simulator for realistic LTE/NR scenarios
- OMNeT++: Discrete event simulator providing the core simulation framework
- VEINS: Vehicular network integration layer
- SUMO: Realistic traffic and mobility patterns for vehicles
- Handover success rate
- Handover latency (milliseconds)
- Packet loss ratio during transitions
- Prediction accuracy (precision, recall, F1-score)
Deep Learning Integration: Seamless integration of PyTorch/TensorFlow models with Simu5G simulation engine
Realistic Scenarios: Vehicle mobility, path loss, interference, and multi-cell coverage variations
Ensemble Comparison: Systematic comparison of single-model vs. ensemble approaches
Comprehensive Analysis: Performance benchmarks across different vehicle speeds and traffic loads
Detailed documentation about the project available in: README_PROACTIVE_HO.md