Traffic-Forecasting-System

Machine learning algorithms for traffic data forecasting and route-finding on the Norwegian road network
Forecast traffic at TRPs, build road-network graphs, and compute traffic-aware routes.

---

Overview

Traffic-Forecasting-System is an open-source system that:

• Reconstructs the Norwegian road network as a graph (nodes = intersections, edges = road links) using NVDB and Trafikkdata API data sources.
• Trains ML models (decision-tree and histogram-gradient-boosting regressors) to forecast traffic volume and average speed at traffic registration points (TRPs).
• Uses those forecasts, plus spatial interpolation (Ordinary Kriging), to evaluate traffic on road segments and produce up to N traffic-aware routes between two points.

Key features

• Ingest Trafikkdata (GraphQL) and NVDB spatial data.
• Modular project-per-database architecture (isolated projects).
• Data cleaning with the additional implementation of MICE and zero-adjusted gamma regression to generate zero data when there is no traffic.
• Distributed training & hyperparameter tuning with Dask (GridSearchCV).
• Route-finding using A* and iterative path diversification, with weights accounting for length, lanes, speed limits and predicted traffic.
• Spatial interpolation of TRP predictions using PyKrige (Ordinary Kriging).

---

Requirements

• Python 3.11+
• PostgreSQL with PostGIS extension
• Dask
• Dask Distributed (for distributed computations)
• Dask-ML
• Access to Trafikkdata API (public GraphQL endpoint) and NVDB data (public).