Recently, Altair unveiled a fascinating study of electric vehicle (EV) adoption across the country. The study demonstrates the potential of using machine learning (ML) to predict and plan for large-scale population-level behaviors.
The analysis identified EV adoption rates down to the county level of 15 states. Altair applied ML to the data from those states and used it to create complete historical datasets to predict EV adoption rates of counties in the country’s remaining 35 states.
Not only did the study provide an in-depth look at the rate of EV adoption, giving policymakers and companies a valuable perspective on the market—but it also showcased the power of Altair’s AI-powered advanced data analytics to derive valuable insights out of complex, limited or incomplete data.
Altair’s data scientists began the process by gathering available information from public domain sources in 15 states from 2019, broken down by county. They identified a series of indicators and variables—almost 75 of them. From those, they selected 30 for modeling purposes on subjects such as the number of charging stations available in the county, households with Internet access, commuting habits, education levels, income, local laws and available EV incentives. Even the 2016 presidential election results were a factor.
The scientists then applied machine learning to the data to help predict and explain the level of EV adoption in the rest of the country.
The county-by-county data was collected and preprocessed through access application programming interfaces (APIs). Altair used R (a programming language and free software platform geared toward statistical computing) to access those APIs and convert the data into CSV files.
The files were then input into Altair’s Knowledge Studio, an ML and predictive analysis tool that visualizes data and generates explainable results without requiring coding. Knowledge Studio was used to prepare the data, create models, and evaluate, validate and score the models. The scored data was then output into a spreadsheet file.
Altair used Panopticon, the company’s data visualization and analytics streaming product, to create visualizations of the data (see the impressive results below).
For states where EV registration data was unknown, Altair’s model predicted that the counties that would see the highest level of EV adoption had populations with three commonalities:
- A relatively high proportion of college-educated residents
- Relatively high median household incomes
- Most commonly lived in densely populated metropolitan areas
This makes intuitive sense as:
- Electric vehicles can be expensive.
- The charging infrastructure needed to run the vehicles is more easily accessed in dense urban centers.
Altair also factored in the effect of local governments and public organizations into its analysis. That influence was overwhelming: about 90 percent of the counties had medium to high numbers of charging stations, and almost half of those jurisdictions were in states that had at least 10 laws and incentives aimed at electric vehicles.
“By applying ML to relevant historical data, we can impute the missing values in that data,” said Sam Mahalingam, Altair’s chief technical officer for HPC/Cloud Solutions. “Now, businesses can use the data to inform strategic decisions, such as where, when, and how to best market electric vehicles. For politicians, the insight can support better policy making, such as considering the merits of investment in charging infrastructure or tax breaks for zero emission vehicles.”
In fact, Altair’s data showed that the availability of charging stations is a crucial variable for EV adoption. In the U.S., there are about 100,000 public chargers currently available—emphasizing the nature of the shortage. The Biden administration’s Infrastructure Investment and Jobs Act—which provides $7.5 billion to start the President’s goal of increasing that number to 500,000 EV chargers nationwide by 2030—aims to significantly close the gap between demand and available infrastructure that this data has highlighted.
“What we can see from the model is that the availability of charging stations is a very important driver,” said Mamdouh Refaat, chief data scientist at Altair to Skip Descant of Government Technology. “You have to have the infrastructure first to encourage people to start buying it.”
For carmakers and regulators alike, data like this can be invaluable to guide business and public policy decision-making alike. Vehicle makers and similar companies can use the information to find and market to a growing number of existing and potential EV buyers. Politicians and policymakers can use the same data to inform their decisions and maximize the use of limited public funds. As the data shows, public and private sector efforts will likely both be targeted at affluent urban populations. But the data could be used to push EV adoption forward by strategically targeting potential avenues for growing the market.
This kind of data isn’t limited to just electric vehicles, either. As ML becomes more of a mainstream data analysis tool, it’s likely to increase attention from lawmakers and businesses seeking to further the objectives they were elected on or gain a competitive advantage in the market. Policymakers could use ML to analyze traffic data to target road infrastructure investments, sift through tax filing data to track down tax evasion and fraud, and match rehabilitation programs in correctional settings to the inmates who would best benefit from them.
A key success element will be explainability—that is, there will be plenty of pressure on data scientists to explain their results, and the underlying algorithms, to decision makers who will not have the same level of expertise—but will also be factoring in external political pressures that might influence what they do with the data.
“AI and machine learning are most valuable when used to support, and not substitute for, human decision making,” according to analysis by McKinsey. “Many public-sector systems are already designed to enable this, such as judicial hearings and public-comment periods around policy decisions. This combination of ‘human plus machine’ can actually often make substantively better decisions than the machine or the human on their own.”
While simpler algorithms can be easier to explain to politicians, they trade off a certain level of accuracy. Conversely, more nuanced and complex algorithms may get “lost in translation,” even if the data they generate is valuable. It’s important to find the right balance: a policy maker needs to be able to provide a rationale for their decisions—which makes explainability so important—while at the same time having confidence in the accuracy and details of the data.
“In undertaking this project, we were eager to showcase Altair’s data analytics capabilities,” said Mahalingam. “But there is also a larger story here. In recent years, AI and big data have generated plenty of negative headlines. Some of these reflect justified concerns over privacy and the consequences of rogue and biased algorithms. However, we shouldn’t ignore how we can use AI and data analytics to create lasting, positive change.”
Read more about efforts to expand the uptake of electric vehicles at Bringing Vehicle Electrification to the Mainstream.