Intelligent Transportation System (ITS) using DataCivet

Background and Objectives

In an intelligent transportation system, traffic prediction is critical. Accurate traffic forecasting can help with route design, vehicle dispatching, and traffic congestion reduction. This study uses the Machine Learning models, specifically, Artificial Neural Networks, Random Forests, and Support Vector Regression algorithms to correctly model traffic flow at different data resolutions and respond to unanticipated traffic incidents as a first step toward addressing these issues. We also investigate several feature selection strategies in order to uncover and better understand the spatio-temporal attributes that have the greatest impact on the model's reliability. DataCivet, IVA’s no-code Graph-Machine Learning (G-ML) platform, automates traffic congestion forecasting from data import and preparation to multiple model development and comparisons, as well as dashboard displays of the forecasts for Delhi Police.

Method

DataCivet is a platform that automates the whole G-ML lifecycle, from data ingestion to dashboard-based visualisation. It is available as a standalone or cloud-based solution. It anticipates traffic congestion by simultaneously building a large number of predictive models, optimising their performance, and comparing prediction metrics to determine which model is best for a specific scenario. We model this context with Graphs, which provide a representational and learning capability that pure machine learning algorithms lack. A user-driven on-demand dashboard is constructed, which displays the ML model output parameters as well as a real-time spatial prediction of traffic congestion.

Results

DataCivet is a generic platform that uses graphs, G-ML, and deep learning models to handle a variety of data-driven challenges in prediction and classification.. It has developed proprietary data preparation techniques that can extract context and include graph-based metrics into the machine learning process. The current condition of traffic is defined by a mix of traffic characteristics such as volume and occupancy in the event of traffic congestion. We evaluate both volume and occupancy because they each give useful information, unlike many previous studies that have simply considered traffic flow when making forecasts, which does not describe the traffic condition uniquely.

Conclusion

In countries all around the world, traffic congestion results in enormous monetary losses. Traffic prediction is more valuable than using real-time traffic data since it allows you to make judgments based on traffic estimates in the near future. Manually constructed models take too long and are incapable of dealing with the volume, diversity, velocity, and validity of data that becomes available in real time. IVA proposes that DataCivet, their Graph-ML platform, be used to solve this challenge. DataCivet can automate the whole prediction lifecycle, from data ingestion to output dashboards, without the need for skilled machine learning people.