Safe Road: How Data Can Make Us Safer through Machine Learning Models
“The ambulance service is simply not providing the levels of service they should -patients are waiting too long and that is putting them at risk.”
Road safety remains one of the great unsolved challenges of the 21st century. While all forms of modern transportation have seen great strides in safety improvement over the course of the past half century, the safety record of the automobile stands out like a sore thumb. The promise of autonomy and better driver assistance technologies are reasons for optimism, but they have yet to mitigate the ugly truth of modern road safety. What we do have in the here and now, however, is a treasure trove of data. With this data comes the possibility to make positive change a reality by extracting new information that could be used to help the government make more informed decisions and allocate resources more efficiently.
It is estimated that fatal and nonfatal crash injuries will cost the world economy approximately $1.8 trillion dollars from 2015–2030. This amount is equivalent to a yearly tax of 0.12% on global GDP. But the consequences are far from being purely financial. Road crashes are the eighth leading cause of death globally for all age groups and the leading cause of death for children and young people 5-29 years of age. This tragic loss of life is rendered even more upsetting by the fact that 66 percent of all fatalities due to EMS delays can be fixed within a four minute window. Helping government agencies allocate road safety resources more effectively could significantly alleviate these unnecessary deaths without requiring major spending bills.
To help provide better resource allocation and safer infrastructure, a team of machine learning enthusiasts from UC Berkeley decided to test on projecting injury related accidents and providing strategic recommendations on allocating resources.
Before the team chose which model to use for their application, they utilized the Mean Squared Error as a metric to ensure they are getting the highest quality results possible. Once their goals had been established, they dove deep into the type of models that would provide cutting edge prediction such as the vector autoregression (VAR), the Prophet, and Object-Oriented Bayesian Time Series (Orbit). They explained how the VAR was used as a baseline model since it was the simplest to implement, and provided reliable results. Furthermore, the Prophet and Orbit models were both really good at predicting seasonality. But when it came down to comparing their MSE; the Orbit came in at the lowest, so they incorporated it into the User Interface. The use of Transfer Learning and Retraining made for the optimization of an accurate and predictable model.
Their final approach provided insight into how to build safer communities by allocating resources using Bayesian Time Series. One of the critical concerns with current individuals is the need for resource allocation to serve more users. Utilizing the Orbit Model, they were able to project data into the future to be able to understand trends, concerns, and issues with current public safety practices. In addition, they utilized seasonality graphs from the Prophet model to identify general, annual, and monthly abnormalities within the data set. Overall, the models provided them with accurate results that build safer communities for the future.
So what do users think of the application? We sat down with the team and they explained who they interviewed, and what feedback they received. They chose to interview local government employees, since oftentimes, safety of our local roads is a local level issue. The test users all thought that the seasonality graphs were useful for determining what time during the year to better allocate public resources, and they all agreed that safety remains the number one issue. This was great feedback, and the team knowing that they are addressing an issue decided to move forward with making this application available to anyone.
They are confident that they are able to build safer communities and hope SafeRoad can be a staple in the public safety space. Safety is their number one priority.
Project By: Christopher Diaz, Clare Lei, Sebastien Nejadnik, Jack Tseng, and Vignesh Sivaprakasam
Industry Mentor: Yael Zheng