A Machine Learning Approach to Understanding Sociodemographic Factors in Electric Vehicle Ownership in the U.S.

Abstract

Electric vehicles (EVs) are rapidly gaining popularity due to their environmental benefits, such as reducing greenhouse gas emissions. Considering the sociodemographic factors that influence the adoption of EVs is essential when developing equitable and efficient transportation policies. This article leverages the National Household Travel Survey (NHTS) 2022 data to analyze the sociodemographic factors influencing the adoption of EVs in the U.S. A binary logistic regression model and three machine learning models were employed to predict EV ownership in the U.S. The results of the regression model suggested that the Pacific division leads in EV adoption, most likely due to legislation and improved infrastructure, while regions such as East South Central suffer from lower EV adoption. The findings indicate that higher household income and home ownership significantly correlate with increased EV adoption. In contrast, renters and rural households exhibit lower adoption rates suggesting an increase in charging facilities in these regions can promote EV adoption. The Random Forest model outperforms others with an accuracy of 82.72%, suggesting its robustness in handling complex relationships between variables. Policy implications include the need for financial incentives for low-income households and increased charging infrastructure in rural and underserved urban areas to promote equitable EV adoption.

1. Introduction

The transportation sector is responsible for a significant increase in greenhouse gas (GHG) emissions in the United States, accounting for approximately 28% of total GHG emissions. Between 1990 and 2022, GHG emissions in the transportation sector increased more in absolute terms than in any other sector [1]. Passenger cars, as well as light and heavy-duty trucks, contribute significantly to pollution [2]. For instance, light-duty vehicles in New York City (NYC) emit 80% of the city’s total transportation emissions [3]. Predictions about the future role of EVs in light-duty vehicle markets show a broad range of possibilities: anywhere from around 10% to nearly 100% of light-duty vehicle sales by 2050 [4,5]. The current transportation system is heavily reliant on fossil fuels, resulting in significant greenhouse gas emissions that contribute to climate change, with low-income populations bearing an uneven share of the resulting health issues, extreme weather events, and other environmental issues. A low-carbon energy transition is vital to mitigating climate change. Still, it can also have energy justice and equity implications for Black, Indigenous, and People of Color (BIPOC), low-income, and other frontline communities [6].

Electric vehicles (EVs) have become a cornerstone of the global transition toward sustainable energy, driven by advancements in lithium-ion batteries that offer higher energy density and lower costs. Their adoption not only reduces greenhouse gas emissions but also supports renewable energy integration through vehicle-to-grid (V2G) technology, which enables bidirectional energy flow. EVs act as mobile energy storage units, stabilizing energy grids while benefiting from cleaner power sources like solar and wind. This synergy between EVs and renewables accelerates transportation electrification and enhances grid reliability. To fully realize their potential, continued innovation in energy storage technologies and supportive policies are essential [7,8,9].

To address these challenges, a variety of measures have gained traction, including efforts to encourage the adoption of vehicles with superior fuel efficiency, the utilization of alternative fuels like biofuels, electricity, and wind energy, as well as regulations in road transport [10]. For instance, EVs are widely recognized for producing zero tailpipe emissions, as no reliance on fossil fuels is required. By eliminating the need for combustion, EVs significantly reduce GHG emissions. Moreover, these vehicles help improve air quality by reducing harmful pollutants such as nitrogen oxides and particulate matter, which are known to cause respiratory issues and contribute to overall air pollution. The shift to electric cars holds the potential to enhance air quality, benefiting both individuals and communities alike. In a study by Malmgren (2016), the environmental advantages of EVs were quantified over a lifespan of 10 years and 120,000 miles of driving [11]. The findings indicated that each electric vehicle generates savings of $5618 in maintenance and fuel costs, along with additional environmental, health, national security, and economic benefits amounting to $6785. Additionally, research by the National Bureau of Economic Research (NBER) highlighted the broader environmental gains of EV adoption, with benefits reaching as high as $3025 in the state of California [12].

Beyond electric vehicles, fuel cell vehicles also play a crucial role in reducing GHG emissions. Jia et al. (2024) introduced a health-oriented energy management strategy (EMS) for fuel-cell hybrid electric buses (FCHEB). The primary objective was to optimize hydrogen fuel consumption, enhance fuel cell durability, and maintain battery thermal health. The researchers gathered environmental data, real-time road conditions, and geographic information using vehicle sensors, Global Positioning System (GPS), and Geographic Information Systems (GIS), which allowed them to formulate the energy management strategy. The study also created a real-world validation model that accounted for factors like terrain, ambient temperature, and driving conditions. The results demonstrated that the proposed EMS could extend battery life by 28.02% and improve overall vehicle efficiency by 8.92% [13].

Moreover, transportation is the second-largest household expenditure, and low-income households face additional challenges due to fixed costs associated with car ownership [14]. The primary challenges to EV adoption, including all-electric and plug-in hybrid electric cars, are the initial purchase price, vehicle range, and charging infrastructure availability [15,16]. The higher cost of EVs has been identified as a more significant barrier to consumers than the EV range [17]. Given the importance of charging for EV adoption, and the large amount being invested in charging infrastructure, it is natural to wonder about the fairness of these investments and deployments of EVs. Despite the environmental benefits, not all social and economic groups in the U.S. adopt EVs equally. The unequal distribution pattern of EVs gives rise to questions over their accessibility and equity of sustainable transportation options. Previous studies investigate various aspects of the equitable transition to electric transportation, as well as discrepancies in access to EV charging infrastructure based on race and income. For instance, Khan et al. (2022) find that environmental and transportation equity disproportionately affects low-income and minority populations [18]. Moreover, home chargers are the most significant and often utilized form of charger in EV adoption and operation, public chargers are critical for people who lack off-street parking and home chargers. Installing home chargers is also more challenging in rental residences because renters are less likely to bear the cost of an upgrade to a home not owned by them, and owners are less likely to bear the cost of a charger not used by them [15]. At-home charging may not be possible for renters, lower-income households, multi-family housing occupants, or those without garages due to installation costs and affordability [4].

The Biden–Harris administration’s Justice40 Initiative has the potential to address such inequity and underinvestment by allocating 40% of the overall benefits of Federal investments to disadvantaged communities that are marginalized, underserved, and burdened by pollution and environmental hazards [19]. The issue of uneven access to EV charging stations also includes the availability of suitable and interoperable charging connectors, fast charging stations, and charging costs. Existing discrepancies between and within advantaged and disadvantaged groups should be closely monitored to prevent increasing imbalances and provide fair access to EV charging facilities for all [20]. This article leverages the 2022 National Household Travel Survey (NHTS) data to analyze the demographic and socioeconomic factors influencing the adoption of EVs in the U.S. Focusing on the type of fuel vehicle runs on variables from the vehicle dataset, the study investigates the correlations between EV ownership and various equity and sociodemographic indicators, including family income, race, number of vehicles in the household, and home ownership, etc. By examining these factors using machine learning models, this study aims to uncover patterns and disparities in EV adoption, contributing to a nuanced understanding of the equitable distribution of this technology.

The motivation for this study stems from the growing urgency to transition to sustainable transportation systems and the need to address inequities in the adoption of EVs. Despite their environmental benefits, EV adoption remains un-even, disproportionately favoring higher-income households and urban areas with better infrastructure. This inequity poses significant challenges to achieving widespread adoption and equitable access to EV technology. By leveraging data from the 2022 NHTS, this research seeks to uncover the sociodemographic and geographic factors driving these disparities, to inform policies that promote equitable access to sustainable transportation solutions.

The scope of this study focuses on understanding the factors influencing EV adoption across the U.S. using data from the 2022 NHTS. The research examines the sociodemographic, economic, and geographic determinants of EV ownership, emphasizing equity and accessibility in sustainable transportation. The thematic focus is on uncovering disparities in EV adoption and providing actionable policy insights to address inequities in access to EV technology and infrastructure. This study contributes to the understanding of EV adoption by leveraging the 2022 NHTS to analyze sociodemographic factors influencing ownership in the U.S. It integrates traditional regression and machine learning models to enhance predictive accuracy and reveal complex relationships. The findings provide insights into the socioeconomic and geographic factors that drive EV usage, highlighting potential areas for policy intervention to promote inclusive access to sustainable transportation.

2. Current State of Knowledge

2.1. Sociodemographic Information

Considering the sociodemographic factors that influence the adoption of EVs is essential when developing equitable and efficient transportation policies. The equitable distribution of EV charging infrastructure has emerged as a critical issue in addressing disparities in access to EVs. Recent studies have underscored significant sociodemographic disparities in the availability and accessibility of public EV charging stations, revealing a complex interplay between income, race, and urban development. For instance, lower-income, Black, and disadvantaged neighborhoods in NYC have fewer EV charging stations, highlighting a lack of equitable distribution [18]. In California, disadvantaged communities often residing in multi-unit dwellings (MUDs), have less access to home charging infrastructure and are more likely to purchase lower-range used EVs. Consequently, these communities should have greater access to public charging [15]. Additionally, public charging stations are more accessible in areas with fewer single-family homes and more commercial areas, regardless of population and EV adoption rates [21].

Similar findings from U.S. studies show that the availability of EV charging stations is not determined by population density [18,22], but is correlated with median household income [18,22,23,24] age [22], percentage of white-identifying population [18], and the presence of highways within a zip code area [15,18]. Extended this analysis, where central residents have better access to medium and quick chargers, showed that higher education levels, income, and private housing are linked to more equitable EV charging access. There are significant correlations between sociodemographic factors like age, education, family structure, and housing type with EV charger accessibility, revealing spatial heterogeneity [25]. Further supporting these findings, Caulfield et al. identified charging inequity between rural and urban areas and among different income groups [26]. There is a complex relationship between sociodemographic characteristics and access to EV charging infrastructure. Funke et al. emphasized that a city’s demographic profile and population density are key indicators for determining the need for charging infrastructure [27].

In contrast, exploring barriers to EV adoption based on consumer perspectives found that in a survey of 733 respondents, financial, technological, and infrastructure challenges are the primary obstacles, with less concern about environmental factors. This suggests that while environmental benefits are acknowledged, practical concerns significantly influence adoption rates [28]. Recent studies, such as those by Hopkins et al. and Soltani Mandolakani and Singleton, emphasize the need for integrating equity considerations into EV infrastructure planning. Hopkins et al. highlighted the necessity of ensuring equitable access to advancements in EV technology to support climate goals. Soltani Mandolakani and Singleton proposed frameworks for measuring and addressing equity and justice in EV infrastructure deployment, advocating for rigorous analysis of accessibility and resource distribution [4,29].

In conclusion, previous studies consistently demonstrate an inequitable distribution of EV charging infrastructure, with underprivileged and minority populations experiencing considerable barriers. To address these inequities, specialized policies and initiatives must be developed, and the diverse requirements of different communities must be considered, to ensure that the benefits of EV adoption are delivered equally. Future research should focus on these issues, using disaggregated data and comprehensive frameworks to provide effective solutions for encouraging equal access to EV charging infrastructure.

2.2. Built Environment

In addition to sociodemographic factors, the built environment is a significant determinant in the adoption of EV infrastructure. Built environment factors may also contribute to the level of EV adoption. Regarding how various infrastructure elements contribute to disparities in transportation access, even seemingly small infrastructure elements, such as pedestrian crossings, can reveal broader disparities in transportation access among residents based on sociodemographic traits [30]. Public charging stations are more easily accessible in places with a higher concentration of commercial areas and fewer single-family houses, according to an analysis conducted by Esmaili et al. on the distribution of EV charging stations in King County, Washington.

A new accessibility measure was used to account for EV competition, aligning with utilitarian and capabilities-based equity theories. The spatial analysis supported a clustered distribution of charging stations [21]. Furthermore, a data-driven approach is used to strategically determine the location and sizes of charging stations in San Francisco. This strategy considers factors such as social equity, EV charging demand coverage, and site development expenses. It aims to provide a comprehensive approach that addresses both infrastructure needs and social equity concerns [31].

Building on identifying spatial disparities in public EV charging infrastructure across the U.S., regions with higher proportions of carless households and fewer vehicles have a greater density of charging stations [20]. This aligns with previous findings that low-income households, often with fewer vehicles, show a positive correlation with charging station density in areas with at least one station. Regarding identifying spatial differences in public EV charging stations throughout the U.S., areas with a larger percentage of families without cars and fewer vehicles overall exhibit a higher concentration of charging stations [20]. This is consistent with earlier research that found a direct relationship between the density of charging stations and low-income households, which often have fewer vehicles. This relationship is observed in locations that have at least one charging station.

While these studies provide valuable insights into the environmental benefits of EV infrastructure adoption, the use of narrow geographical and demographic data often limits them. For instance, studies that focus on a specific set of regions, characterized by technical constraints on size or continuity, may not reflect the broader trends seen in other areas. Similarly, analyses based on census tract-level data can be problematic, as these tracts may not fully represent a larger state or national context. Tracts can also contain diverse populations with varying socio-economic, demographic, and social characteristics, which may distort generalizations. For example, a tract may combine both affluent and impoverished areas, making it appear as if it is a middle-income zone. To address these challenges, this study utilizes the 2022 NHTS dataset, the largest national transportation dataset in the U.S. By leveraging such comprehensive data, this research offers a more accurate, generalized model that can be easily applied to different states or countries.

3. Methodology and Data

3.1. Data

The NHTS is the largest and most valid national transportation-related dataset in the U.S. The NHTS is a nationally representative data source for daily local and long-distance passenger travel administered by the United States Department of Transportation (USDOT) [32]. The survey was conducted nine times: in 1969, 1977, 1983, 1990, 1995, 2001, 2009, 2017, and 2022. It provides a comprehensive record of how travel behavior has evolved with changes in demographics, economics, and culture [33]. The NHTS includes four core linkable tables, each with one record per household, person, trip, and vehicle. The data include variables at the household, person, vehicle, and trip level. In this study, vehicle and household-level datasets are used. The household dataset describes the household characteristics of each respondent. The vehicle dataset describes the vehicle characteristics of each vehicle in the household [34].

To integrate the vehicle and household datasets from NHTS 2022, the authors perform a merge operation using the common HOUSEID identifier in the dataset. This process involved utilizing the inner join function from the “dplyr” package in R software (version 2024.04.2) [35]. By merging the HOUSEID column, a unified dataset that consolidates vehicle-specific data with corresponding household attributes is created, facilitating a comprehensive analysis of travel behavior and patterns. Moreover, the appropriate survey weights were applied to the merged dataset to ensure the NHTS 2022 data accurately represents the national population. Using the survey package in R software, a survey design object that incorporated the weights provided in the dataset was created. This weighting process adjusts for sampling bias and allows for a more reliable and generalizable analysis of travel behavior and trends.

This dataset contains 14,684 responses from household respondents. The weighted data includes a population of 232,837,104. Moreover, the vehicle dataset includes a question about the type of fuel the vehicle runs on. One of the responses is “electric only”. Therefore, anyone who answered electric only to the fuel type for their vehicle is considered an EV owner in this study. From the sample size, 186 respondents were considered EV owners, and the weighted population includes 2,898,975 EV owners (1.25% of the total population).

Table 1. Vehicle and Household Characteristics of the NHTS 2022 Data.

Variable	Population (Unweighted)	Population (Weighted)	EV Owners (Weighted)
Sample Size	14,684	-	186
Weighted population	232,837,104	-	2,898,975
	Vehicle Characteristics
Census division classification for home address
New England	5.26%	4.69%	5.14%
Middle Atlantic	10.11%	10.83%	6.67%
East North Central	17.32%	15.88%	7.62
West North Central	7.93%	7.39%	4.26
South Atlantic	19.57%	19.94%	21.56%
East South Central	5.98%	6.2%	2.16%
West South Central	10.88%	11.56%	4.75%
Mountain	8.20%	7.82%	4.97%
Pacific	14.76%	15.69%	42.86%
Number of drivers in the household
Zero Driver	0.25%	0.36%	0.00%
One driver	22.68%	22.49%	15.30%
Two drivers	59.64%	54.89%	65.68%
Three or more drivers	17.44%	22.25%	19.02%
Vehicle ownership
1 vehicle	17.86%	18.15%	12.85%
2 vehicles	43.57%	41.53%	51.87%
3 or more vehicles	38.57%	40.32%	35.27%
Household income
Less than $25K	7.55%	9.16%	1.52%
$25K–$49.9K	14.75%	15.68%	2.41%
$50K–$99.9K	33.06%	32.49%	20.54%
$100K and above	44.65%	42.67%	75.53%
Household size
1 Person	17.41%	16.70%	13.62%
2 persons	45.72%	37.21%	34.03%
3 persons	14.91%	18.29%	28.66%
4 persons and more	21.95%	27.80%	23.69%
Household in urban/rural area
Urban	75.48%	76.51%	86.4%
Rural	24.52%	23.49%	13.6%
Urban area size where home address is located
50,000–199,999	9.67%	10.16%	16.38%
200,000–499,999	10.86%	10.93%	7.23%
500,000–999,999	9.45%	9.23%	5.63%
1,000,000 or more	36.86%	37.48%	54.49%
Not in urbanized area	33.16%	32.21%	16.27%
Count of workers in household
No worker	29.45%	27.23%	15.68%
One worker	32.26%	32.58%	27.41%
Two workers	31.11%	30.81%	48.27%
Three workers and more	7.18%	9.38%	8.65%
Whether home owned or rented
Owned with mortgage/loan	53.05%	52.83%	67.66%
Owned (no mortgage)	29.35%	25.57%	23.57%
Rented	16.32%	20.07%	6.73%
Occupied without payment	1.28%	1.54%	2.04%
Type of home
One-family detached	78.91%	76.35%	83.60%
One-family attached (townhome, condo)	7.70%	7.72%	10.38%
Building with 2 or more apartments	10.13%	11.71%	4.43%
Mobile home	3.02%	3.89%	1.60%
Boat, RV, van, etc.	0.25%	0.33%	0.00%
Race of household respondent
White	87.09%	79.94%	69.62%
Black or African American	5.59%	8.59%	2.67%
Asian	4.23%	5.94%	21.04%
Other	3.08%	5.52%	6.66%

shows the vehicle and household characteristics of the merged dataset and the weighted percentage.

3.2. Methodology

A binary logistic regression model and three machine learning models, including Naïve Bayes, Support Vector Machines, and Random Forest, are employed in this study to make a prediction on electric vehicle ownership in the U.S. using NHTS 2022 data.

While each of these models has distinct characteristics useful for binary classification tasks, some disadvantages also accompany them. For instance, the binary logistic regression is straightforward, interpretable, and efficient for linearly separable data, however, this model is only limited to linear relationships. The Naïve Bayes model assumes that features are independent given the target class and can handle cases with categorical features effectively. However, this model has a strong independence assumption which reduces the performance if features are highly correlated. In addition, the support vector machine SVM is powerful and works well with a clear margin of separation between classes. Its flexibility in choosing different kernel functions (linear, polynomial, radial basis, etc.) allows it to capture complex decision boundaries. The disadvantage of this model is that it can be computationally intensive, especially with large datasets such as the NHTS 2022 data. Moreover, the Random Forest combines multiple decision trees to improve robustness and reduce overfitting, making it effective for data with complex patterns. It is particularly good at handling noisy data and capturing non-linear relationships. This model is considered less interpretable than the simpler models above and requires careful tuning.

The data had already been pre-processed since the vehicle and household-level data were merged at the very beginning and based upon a common identifier. Selected features or independent variables include Census divisional classification, number of drivers in the household, household income, home ownership, urban/rural classification, number of workers in the household, etc. The dependent variable, EV ownership, was recoded into binary format differentiating electric-only vehicles from others.

Since EV owners are only 1.25% of the total population, and EV users are in a minority in the sample dataset, the Random Over-Sampling Examples (or ROSE) were used to balance the imbalanced dataset and create synthetic balancing of classes for robust training of models. Subsequently, the dataset was balanced using this method, and then it was split into training and test sets, maintaining the level of factors identical in both subgroups in this study. First, a binary logistic regression was applied to balanced training data to see the relation of predictor variables with the likelihood of EV ownership. A variance inflation factor (VIF) measures the degree of multicollinearity in regression analysis. Multicollinearity occurs when a multiple regression model has a correlation between numerous independent variables. This can have a negative impact on regression results [36]. The formula for VIF is demonstrated in Equation (1):

$$\mathrm{VIF}={\displaystyle \frac{1}{1-R_i^2}}$$

(1)

where $R_i^2$ is the unadjusted coefficient of determination for regressing the ith independent variable on the remaining ones. Subsequently, the Naïve Bayes, SVM, and Random Forest models were trained on the balanced training data and evaluated key metrics of performance on the testing set: accuracy, precision, recall, and F1-score. The “Accuracy” metric is good for balanced datasets, while the “Precision and Recall” metrics offer more insights into imbalanced datasets. In addition, the “F1-score” metric balances precision and recall, making it helpful when there’s a need to balance false positives and false negatives. Some alternative metrics could offer additional perspectives. For example, the Area Under the Receiver operating characteristic (ROC) Curve (AUC-ROC) metric captures the trade-off between true positive rate (sensitivity) and false positive rate across threshold values. AUC-ROC is beneficial in evaluating how well the model separates the classes [37]. In addition, the Matthews Correlation Coefficient (MCC) metric is particularly useful for binary classification with imbalanced classes, as it accounts for true and false positives and negatives in a balanced way [38].

3.2.1. Binary Logistic Regression Model

A binary logistic model was used in this study to investigate the EV adoption behavior in the U.S. The variable “VEHFUEL” was used as a dependent variable which is about “Type of fuel the vehicle runs on”. The levels of the variables were merged into two categories: Electric and non-electric. Independent variables are users’ sociodemographic information, including household income, employment status, driver status, etc. First defined in the 1960s, the logistic regression model (LR) is widely used to deal with the discrete choice problem [39]. Binary logistic regression determines the impact of multiple independent variables presented simultaneously to predict membership of one or other of the two dependent variable categories. In logistic regression, a logistic transformation of the odds (referred to as logit) serves as the depending variable (Equation (2)): [40]

$$\log \left(odds\right)=logit\left(pP\right)=\ln \left({\displaystyle \frac{P}{1-P}}\right)$$

(2)

By taking the aforementioned dependent variable and incorporating it into a regression equation with the independent variables, a logistic regression model is obtained (Equation (3)):

$$logit\left(P\right)=\alpha +b_1{x}_1+b_2{x}_2+b_3{x}_3+\cdots$$

(3)

As in least-squares regression, the relationship between the logit(P) and x is assumed to be linear. In the Equation (4), P can be calculated with the following formula [40] where:

P = the probability that a case is in a particular category,

exp = the exponential function,

a = the constant (or intercept) of the equation and,

b = the coefficient (or slope) of the predictor variables.

$$P={\displaystyle \frac{\exp \left(\alpha +b_1{x}_1+b_2{x}_2+b_3{x}_3+\cdots \right)}{1+\exp \left(\alpha +b_1{x}_1+b_2{x}_2+b_3{x}_3+\cdots \right)}}$$

(4)

3.2.2. Naïve Bayes Classifier

The Naïve Bayes classifier is a popular supervised machine learning algorithm used for classification tasks such as text classification. It belongs to the family of generative learning algorithms, which means that it models the distribution of inputs for a given class or category. This approach is based on the assumption that the features of the input data are conditionally independent given the class, allowing the algorithm to make predictions quickly and accurately [41]. Equation (5) shows that Bayes theorem provides a way of computing posterior probability P(c|x) from P(c), P(x) and P(x|c).

$$\begin{array}{c}P\left(c|x\right)={\displaystyle \frac{P\left(c|x\right)P\left(c\right)}{P\left(x\right)}}\\ P\left(c|x\right)=P\left(x_1|c\right)\times P\left(x_2|c\right)\times \dots \times P\left(x_n|c\right)\times P\left(c\right)\end{array}$$

(5)

P(c|x) is the posterior probability of class (c, target) given predictor (x, attributes).

P(c) is the prior probability of class.

P(x|c) is the likelihood which is the probability of the predictor given class.

P(x) is the prior probability of the predictor.

3.2.3. Support Vector Machines (SVM) Model

Support vector machine (SVM) analysis is a popular machine learning tool for classification and regression, first identified by Vladimir Vapnik and his colleagues in 1992 [42]. SVM regression is considered a nonparametric technique because it relies on kernel functions. Thus, the classification function can be denoted by Equation (6) [43].

$$f\left(x\right)=sign[{\displaystyle \sum }_{i=1}^n\left(a_i{Y}_j\times k\left(x,x_i\right)\right)+b]$$

(6)

where c is the offset from the origin of the hyper-plane; n presents the number of the independent variables; α_i defines the positive constant; and k (x, x_i) is the kernel function [44].

3.2.4. Random Forest Model

Random forest is a commonly used machine learning algorithm, trademarked by Leo Breiman and Adele Cutler, that combines the output of multiple decision trees to reach a single result. Its ease of use and flexibility have fueled its adoption, as it handles both classification and regression problems. To calculate the factor impurity belonging to each category, the Gini-Index is used to select the factor. It can be computed by Equation (7) [44].

$${\displaystyle \sum }{{\displaystyle \sum }}_{j\ne i}\left({\displaystyle \frac{f\left(Y_i,T\right)}{\left|T\right|}}\right)\left({\displaystyle \frac{f\left(T_j,T\right)}{\left|T\right|}}\right)$$

(7)

To enhance the reproducibility of the proposed methodology and provide clear guidance to practitioners, a pseudo code outlining the workflow is presented in

Table 2. Pseudo Code: Electric Vehicle (EV) Adoption Analysis.

# Define Required Variables INPUT:  -Dataset (D): Preprocessed National Household Travel Survey (NHTS) data.  -Independent Variables (X):    -X1: Census division;    -X2: Number of drivers in the household;    -X3: Household Income;    -X4: Whether home owned or rented;    -X5: Household in urban/rural area;    -X6: Count of workers in household.  -Target Variable (Y): EV Ownership (Binary: 1 = Yes, 0 = No). OUTPUT:  -Model Performance Metrics: Accuracy, Precision, Recall, F1-Score;  -Insights: Key predictors influencing EV ownership. # Load and Preprocess Dataset Step 1:  Load Dataset (D):  -Normalize numerical variables (e.g., X1: Census division);  -Encode categorical variables (e.g., X2, X3) using one-hot encoding. # Split Data Step 2:  Divide Dataset (D) into:   -Training Set: 80% of the data;   -Testing Set: 20% of the data. # Define Models and Parameters Step 3:  -Model 1: Naïve Bayes;  -Model 2: Random Forest;  -Model 3: Support Vector Machine (SVM). # Train Models Step 4:  For each model (M1, M2, M3):   -Train the model using the training set;   -Optimize hyperparameters (if applicable). # Evaluate Models Step 5:  For each model (M1, M2, M3):   -Use the testing set to calculate performance metrics:    -Accuracy;    -Precision;    -Recall;    -F1-Score. # Analyze Results Step 6:  -Compare performance metrics across models;  -Identify the model with the best performance;  -Analyze the importance of each independent variable (X1 to X6). # Output Results Step 7:  -Visualize model performance metrics;  -Provide insights for practitioners:  -Key predictors of EV adoption;  -Policy recommendations to improve EV adoption equity.

. This pseudo-code systematically describes the step-by-step process employed in this study, from data preprocessing to model evaluation and result interpretation.

4. Results

4.1. Descriptive Analysis

Table 1 provides a comparative detail of the vehicle and household characteristics between the total population and EV owners all from the 2022 NHTS. Data showed different characteristics between EV owners and the general population. A weighted population sample includes 2,898,975 EV owners out of a total population of 232,837,104. By geographic region, EV ownership varies widely, with the highest percentage within the Pacific division at 42.86% and the lowest within the East South Central at 2.16%. The percentages of households with multiple drivers and multiple vehicles owning EVs are correspondingly higher. 65.68% of owners have 2 drivers, 51.87% have 2 vehicles, and 75.53% have household incomes above $100,000 compared with the general population at 42.67%. Although household size does not exhibit great disparities, EV ownership is slightly higher in two-person households. Urban households stand a better chance of owning EVs at 86.4% compared to rural ones at 13.6%. Besides, EV owners are drawn from household owners: those under the mortgage, specifically, constitute 67.66% of EV owners. Racial demographics show a greater rate of EV ownership among Asian respondents at 21.04%.

4.2. Binary Logistic Regression Model Results

The results of the binary logistic regression model for important factors affecting the likelihood of U.S. household EV ownership are demonstrated in

Table 3. Binary Regression Model Results.

		Estimate	Std. Error	z Value	Pr(>|z|)
(Intercept)		−4.61	0.63	−7.32	0.00	***
Census division	Middle Atlantic	−0.17	0.41	−0.41	0.68
	East North Central	−0.40	0.40	−1.00	0.32
	West North Central	−0.65	0.52	−1.25	0.21
	South Atlantic	0.09	0.36	0.24	0.81
	East South Central	−0.99	0.66	−1.49	0.14
	West South Central	−0.32	0.43	−0.74	0.46
	Mountain	0.18	0.41	0.43	0.67
	Pacific	0.96	0.34	2.79	0.01	**
Number of drivers in the household		−0.43	0.13	−3.44	0.00	***
Household Income	$25,000 to $49,999	−0.25	0.63	−0.40	0.69
	$50,000 to $99,999	0.25	0.54	0.46	0.64
	$100,000 and above	1.51	0.52	2.88	0.00	**
Whether home owned or rented	Owned (no mortgage)	0.09	0.18	0.52	0.60
	Rented	−0.82	0.29	−2.82	0.00	**
	Occupied without payment	−0.38	1.02	−0.37	0.71
Household in urban/rural area	Rural	−0.57	0.23	−2.45	0.01	*
Count of workers in household		0.15	0.10	1.52	0.13

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 (Dispersion parameter for binomial family taken to be 1). Null deviance: 14,248 on 10,278 degrees of freedom; Residual deviance: 10,862 on 10,243 degrees of freedom. AIC: 10,934—Number of Fisher Scoring iterations: 15.

. The model includes the number of drivers in the household, Census division, household income, home ownership, urban/rural classification, and count of workers in the household, which have a significant impact on EV ownership. After conducting the binary regression model, a backward regression was conducted to remove the best explanatory variables for the model [45]. Moreover, for this study, unweighted data were chosen for the regression analysis to minimize any potential bias resulting from these differences in weight adjustment, since changing weights per the methodology employed in the NHTS data expansion may result in disparities [46]. This choice is predicated on the idea that there may be biases associated with employing weights that are either improper or unadjusted, outweighing the advantages of a weighted method.

The number of drivers has a highly significant negative effect on the possibility of owning an EV (−0.43, p < 0.001), meaning that when the number of drivers increases, the possibility of owning an EV decreases. Higher household income is positively related to EV ownership, with the highest income bracket ($100,000 and above) showing a positive, significant effect (Estimate = 1.51, p < 0.01). This means that wealthier households are more likely to own EVs. Renters are much less likely to own an EV compared to those owning their home with a mortgage (Estimate = −0.82, p < 0.01). Rural households are much less likely to own an EV relative to urban households (Estimate = −0.57, p < 0.05). This implies that the ownership of EVs is more common in urban regions.

4.3. Variance Inflation Factor (VIF)

VIF is the diagnostic test statistic for multicollinearity in regression models. Generally, it is concluded that a VIF value over 10 shows substantial multicollinearity, though some researchers would consider one over 5. When VIF is higher than 10, there is significant multicollinearity that needs to be corrected [36]. The results of the VIF for the variables in the binary logistic regression model are presented in Figure 1. As presented in Figure 1, all the VIFs are lower than 10, which means no multicollinearity among them exists.

4.4. Machine Learning Models Results

The results of the models can be interpreted based on the metrics provided in Figure 2. Accuracy, Precision, Recall, and F1-Score. The proportion of true results (both true positives and true negatives) among the total number of cases examined is accuracy. The proportion of true positive results in the predicted positive cases is precision. The proportion of true positive results in the actual positive cases is recall, and the harmonic mean of Precision and Recall is F1-score. It provides a balance between Precision and Recall.

The results show that the Random Forest model has the highest accuracy (82.72%), followed by SVM (71.58%), and Naïve Bayes (70.49%). This indicates that the Random Forest model is the best at correctly predicting both EV and non-EV ownership overall. All models predict EV ownership with high precision (~99.4%), indicating near-perfect precision. This implies that false positives (predicting an EV when it is not) are uncommon. The Random Forest model has the highest recall (83.05%), followed by SVM (71.63%) and Naïve Bayes (70.52%). High recall indicates that the Random Forest model is good at identifying actual EV owners and has fewer false negatives (missing actual EV owners).

The Random Forest model has the greatest F1-Score (0.9047), indicating a great combination of precision and recall. The SVM model (0.8327) and Naïve Bayes (0.8251) had lower F1-Scores, indicating less balance in comparison.

5. Discussion

The findings of this study, based on the NHTS 2022 data, highlight significant sociodemographic and regional variations in the use of electric vehicles in the U.S. The findings of the descriptive analysis, binary logistic regression model, and machine learning models all highlight the multifaceted nature of EV ownership and how several factors influence this emerging trend.

The Pacific has the highest percentage of EV ownership (42.86%), while the East South Central has the lowest (2.16%). Such a geographical variation would imply that regional regulations and infrastructure availability, as well as environmental policies and approaches, would have a substantial impact on EV adoption rates. Furthermore, the results of this study are in line with the previous studies [15,19,20,21] and indicate that households with higher incomes, and multiple vehicles are more likely to own EVs. For instance, households earning more than $100,000 account for 75.53% of EV-owning families, highlighting the issue of the high cost of owning an EV. More households living in urban areas possess EVs than rural families, which could be attributed to improved charging infrastructure and lower commute times in metropolitan regions. Moreover, more than half (54.49%) live in urban areas with a population of 1,000,000 or more. The higher ownership among households with a mortgage (67.66%) suggests that financial stability and homeownership are conducive to EV adoption, likely due to the feasibility of installing home charging stations. Moreover, most EV owners are living in one-family detached homes.

The binary logistic regression model provides deeper insights into the sociodemographic factors influencing EV ownership. The VIF analysis reveals no significant multicollinearity among the variables included in the binary logistic regression model. This ensures that our regression results are reliable, allowing us to confidently assess the effects of specific predictors on EV ownership. The results of the binary logistic regression model indicate that the Pacific Census division has a significant number of households with EV ownership. This is likely due to more supportive policies, better infrastructure, and higher environmental awareness in these regions.

A higher number of drivers in a home was related to a lower likelihood of owning an EV. This is most likely because multi-driver households prefer conventional vehicles due to their convenience and longer range. Higher household income is positively connected to EV ownership, indicating that financial barriers to EV adoption persist, as higher-income households can easily purchase EVs and home charging infrastructure. Home renters are less likely to own an electric vehicle than homeowners, particularly those with a mortgage. This resolves the issue of tenants’ access to charging infrastructure. Urban families are more likely to own an electric vehicle than rural households. This gap might be explained by the increased availability of charging infrastructure in cities, or, more broadly, shorter commuting distances.

The machine learning models—Naïve Bayes, Support Vector Machines (SVM), and Random Forest—further validate the regression findings and provide comparative performance metrics. It was found that the Random Forest model outperformed the other models in terms of accuracy (82.72%), precision (99.34%), recall (83.05%), and F1-Score (90.47%). Bar graphs have been made regarding each, demonstrating clearly how each model performed in comparison to the others. Based on these findings, it can be suggested that the Random Forest model will be effective in forecasting EV ownership, making it valuable for policy and strategic choices on how to increase electric vehicle adoption.

6. Conclusions

To analyze the factors influencing EV ownership in the U.S., the NHTS 2022 household and vehicle datasets were integrated together in this study. Moreover, the appropriate survey weights were applied to the merged dataset to ensure the NHTS 2022 data accurately represents the national population. The weighted data includes a population of 232,837,104 and 2,898,975 EV owners (1.25% of total population). A binary logistic regression model and three machine learning models, including Naïve Bayes, Support Vector Machines, and Random Forest, were employed in this study to make a prediction on electric vehicle ownership in the U.S. using NHTS 2022 data. Selected features or independent variables include Census divisional classification, number of drivers in the household, household income, home ownership, urban/rural classification, number of workers in the household, etc. The dependent variable, EV ownership, was recoded into binary format differentiating electric-only vehicles from others.

As previously discussed in the review of the literature and the results presented in this study, in order to promote EV adoption in low-income households, key strategies include: implementing targeted financial incentives like increased rebates and tax credits, enhancing access to charging infrastructure in low-income communities, conducting community-based education and outreach programs, and collaborating with local organizations to address concerns and tailor solutions to specific needs. The results of the regression model suggested that the Pacific division leads in EV adoption, most likely due to legislation and improved infrastructure, while regions such as East South Central suffer from lower EV adoption. This would indicate the importance of regional policy initiatives for EV adoption. EV ownership is associated with higher household income and ownership. Financial incentives and support for low-income households and renters might help to improve these problems. Moreover, the results (in line with previous studies [15,19,20,21]) indicate that there are fewer EVs on rural roadways than in urban areas, suggesting an increase in charging facilities in rural regions can promote EV adoption. Moreover, machine learning model results suggested that the Random Forest model outperforms other models in predicting EV ownership, indicating its robustness in handling complex relationships between variables. Overall, subsidies and financial incentives to low-income households and renters to help cover initial expenses, increasing charging infrastructure, particularly in rural and underserved urban areas, and ensuring that everyone, particularly minority and underprivileged communities, has equitable access to EV technology and infrastructure can help to achieve equitable EV adoption.