AI Routing Lab

Predictive route selection by latency/jitter using machine learning to achieve prediction accuracy >92%

Active Started: November 2025

Project Overview

AI Routing Lab is a research project focused on developing machine learning models for predictive route selection in CloudBridge network infrastructure. The project aims to achieve >92% accuracy in predicting latency and jitter for optimal route selection.

The project integrates with quic-test for model validation on real QUIC traffic and with CloudBridge Relay for real-time routing optimization.

Key Objectives

Accuracy >92%

Develop ML models for latency and jitter prediction with R² > 0.92 accuracy

quic-test Integration

Model validation on real QUIC traffic through integration with quic-test

Route Optimization

Real-time route selection optimization through integration with CloudBridge Relay

Current Status

Completed

  • Created LatencyPredictor and JitterPredictor models with ensemble architecture (Random Forest + Gradient Boosting)
  • Implemented RoutePredictionEnsemble for combining latency and jitter predictions
  • Integrated FeatureExtractor from cloudbridge-ai-service for feature extraction
  • Created laboratory framework for experiments
  • Organized reports structure by year, month, and version

In Progress

  • Integration with quic-test for metrics collection and validation
  • Adapting time-series models (LSTM, ARIMA, Prophet) for latency prediction
  • Developing API for CloudBridge Relay integration

Technical Details

Models

  • • LatencyPredictor (Ensemble: RF + GB)
  • • JitterPredictor (Ensemble: RF + GB)
  • • RoutePredictionEnsemble
  • • FeatureExtractor (Time, Statistical, Domain features)

Technologies

Python 3.11+ scikit-learn TensorFlow Random Forest Gradient Boosting quic-test

Performance Metrics

  • • R² Score: >0.92 (target)
  • • Inference Time: ~2ms per prediction
  • • Throughput: 1000+ predictions/sec
  • • Ensemble reduces variance by 15-25%

Model Accuracy Over Training Time

* Target accuracy: R² >92%, inference time <10ms

Model Comparison

* Ensemble architecture reduces variance by 15-25%

Development Status

  • • Models Created: 3
  • • Experiments Ready: 2
  • • Version: 1.1
  • • Status: Active Development

Related Projects & Technologies

Related Research