Browsing by Author "Chan, Travis"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Crash Rate Prediction from Traffic Volume Data using AI
    Chan, Travis; Hossain, Syeda; Le, Jonathan; Asam, Arham; Khadka, Devanshu (Virginia Tech, 2024-12)
    In today's fast-paced, technology-driven world, we're generating more transportation data than ever before. This data offers opportunities to making roads safer and more efficient, but is often hard to take advantage of. Our client, Dr. Mohamed Farag is a researcher in the Center for Sustainable Mobility (CSM) at the Virginia Tech Transportation Institute, a research institute whose work contributes to the advancement of the transportation industry. To address this challenge, we have developed a user-friendly web application that harnesses machine learning to predict crash rates based on traffic volume data. We have developed a web application that allows users to use machine learning models to predict crash rates for roads. It is comprised of four main components: a frontend interface, a backend server, an API, and offline machine learning model development using Google Colaboratory. Administrators have additional privileges, such as managing machine learning models through the Model Management section. They can upload new models, specify model details like names and attributes, and monitor existing models via the Model List page, which displays all models along with their creation dates and statuses. We've implemented secure user authentication on the frontend using JWT tokens for login and sign-up processes. The Home Page presents users with a tabular view of past predictions, allowing them to see the date, model used, and results, as well as the option to add new predictions. Our backend architecture features a Next.js server for the web backend and a FastAPI server for the machine learning backend. The web backend handles user authentication, prediction collections, and model management, while interfacing with the FastAPI ML backend to generate predictions. To ensure quality and reliability, we've conducted extensive testing and evaluation, including machine learning model testing, model evaluation, and client assessments. We recognize that further work is needed to finalize the product. This report outlines our plans for the remainder of the semester and proposes ideas for future enhancements beyond the current project scope—all aimed at making our roads safer through data-driven insights.