Trends in Swiss Passenger Vehicles Based on Machine Learning Segmentation

Abstract

Road transport represents a major contributor to air pollution, energy consumption, and carbon dioxide emissions in Switzerland. In response, stringent emission regulations, penalties for non-compliance, and incentives for electric vehicles have been introduced. This study investigates how these policies, along with shifting consumer preferences and vehicle design advancements, have influenced the composition of the Swiss new passenger car fleet. Using machine learning techniques, we segment passenger vehicles to analyze trends over time. Our findings reveal a decline in micro and small vehicles, alongside an increase in lower- and upper-middle-class vehicles, sport utility vehicles, and alternative powertrains across all segments. Additionally, steady increases in vehicle width, length, and weight are observed in all classes since 1995. While technological advancements led to reductions in energy consumption and carbon dioxide emissions until 2016, an increase has since been observed, driven by higher engine power, greater vehicle weight, and changes in certification schemes.

1. Introduction

In the context of a global energy and environmental crisis, the transport sector is once again in the spotlight due to its heavy reliance on fossil fuels. Notably, the transport sector accounts for more than a third of global CO₂ emissions across all end-use sectors [1]. The rebound following the COVID-19 pandemic led to a 3% increase in transport-related CO₂ emissions in 2022 compared to the previous year [2]. Among all transport modes, the road transport of passenger and goods accounts for around two-thirds of transport-related energy consumption and CO₂ emissions [3]. Furthermore, although new vehicles are subject to stringent emission regulations, tailpipe emissions of particulate matter (PM), nitrogen oxides (NOx), and ozone (O₃) from road vehicles remain a major source of outdoor air pollution, especially in urban areas. These emissions were linked to 385,000 premature deaths worldwide in 2015 [4].

In Switzerland, the transport sector was the largest energy consumer in 2022, accounting for 36% of total end-use energy consumption, primarily driven by fossil fuels (31% gasoline, 40% diesel, and 22% aviation fuel) [5,6]. Passenger cars alone were responsible for 72.3% of the 13.6 million tons of CO₂-equivalent emissions from road transport [7]. Therefore, in order to achieve the CO₂ targets of the Swiss Energy Strategy 2050 [8] and improve overall air quality, Switzerland’s road transport system must undergo a rapid transformation. The adoption of alternative powertrain systems, such as battery electric vehicles (BEVs) and hydrogen fuel cell electric vehicles (FCEVs), along with renewable energy carriers (e.g., synthetic fuels), is key for driving this transition. In 2023, hybrid electric vehicles (HEVs), plug-in hybrid electric vehicles (PHEVs), and BEVs accounted for 57% of new passenger car sales in Switzerland, highlighting the importance of understanding how these trends are shaping the vehicle fleet [9].

Extensive research has explored the potential of alternative powertrains and energy carriers, analyzing their environmental benefits (e.g., [10,11,12,13]) and developing pathways for fleet decarbonization and electrification (e.g., [14,15,16]). However, a significant gap remains in understanding the temporal evolution of vehicle fleet characteristics, particularly in light of the rising vehicle size and weight [17,18,19], advances in engine capacity and fuel efficiency [20,21], and the increasing share of alternative powertrain vehicles [22,23].

Previous studies have shown that technological improvements in fuel economy are often offset by the growing size and power of vehicles. In [24], the authors reported that between 1985 and 1997 in the Netherlands, advances in fuel consumption were offset by increases in vehicle weight and engine capacity. Similarly, studies analyzing the new vehicle fleet in the UK [25], US [26], Sweden [27] and Europe [28] found that the demand for more powerful vehicles with enhanced consumer amenities neutralized much of the technological progress in improving fuel economy up to 2006. The two studies extending beyond 2006 observed that a trend towards downsizing in vehicle weight and engine capacity, coupled with a stabilization in consumer demand for additional features, enabled technological advancements to result in meaningful reductions in fuel consumption until 2015 [27,28]. These findings suggest that while technological progress has been made in improving fuel economy, consumer preferences for larger, more powerful, and feature-heavy vehicles have historically limited the full potential of these advancements.

However, most previous studies consider the entire passenger vehicle fleet, without distinguishing between different vehicle classes or powertrain types, and do not cover the growing market share of electric vehicles. A recent study [18] extends the analysis of vehicle weight, dimensions, and engine capacity and power trends up to 2023, confirming the continued trend toward larger and heavier vehicles across multiple market segments. Their findings align with our results, emphasizing that despite technological advancements, vehicle mass and size have remained on an upward trajectory, which has implications for both energy consumption and emissions. Recent data indicate that global electric vehicle sales increased by 25% in 2024 compared to the previous year, reaching approximately 11 million vehicles globally [29]. Despite this significant shift towards electrification, the trend towards larger and heavier vehicles persists, potentially offsetting gains in energy efficiency. In fact, it has been reported that every 1% increase in vehicle weight leads to an approximately 1% increase in electricity consumption for electric vehicles [30]. Furthermore, the increase in electric vehicle battery capacity, which is often linked to the desire for a greater driving range, further exacerbates this issue. A 10 kWh increase in battery capacity, for example, leads to a 150 kg increase in vehicle mass and a corresponding rise in energy consumption of 0.7–1.0 kWh/100 km [31]. As the demand for a longer range grows, so too does the weight and resource demand of these vehicles, potentially diminishing the energy efficiency benefits brought by electrification. Additionally, recent data from the International Energy Agency highlight that the share of smaller electric vehicle models is declining, with SUVs and large vehicles now comprising the majority of available electric models. This shift mirrors the broader trend observed in the conventional vehicle market, where larger, heavier vehicles have become dominant. As a result, the ongoing transition to electric mobility may not fully compensate for the continued growth in vehicle size and weight, limiting the potential energy savings that could otherwise be achieved.

This study addresses the identified gaps by providing a detailed temporal analysis of vehicle characteristics across various vehicle classes and powertrain types, examining trends in size, weight, engine power, efficiency, and CO₂ emissions within each vehicle class, and analyzing energy consumption and range trends for electric vehicles. These contributions offer a detailed understanding of how technological advancements and consumer preferences impact different vehicle categories over time.

A key aspect of our approach lies in the segmentation of vehicles into distinct classes and sub-classes based on attributes such as size, weight, engine power, number of seats, and drivetrain type. While traditional vehicle segmentation often relies on subjective expert judgment, in previous works, we have presented various machine-learning-based methodologies to objectively segment vehicles based on key dimensions and characteristics. In particular, in [17], we compared the performance of K-means and Fuzzy C-Means (FCM) algorithms for segmenting the fleet of new passenger cars registered in Switzerland in 2018. Further, in [32], we demonstrated the effectiveness of semi-supervised FCM and other state-of-the-art classification methods—including k-nearest neighbors (kNN), AdaBoost (Ada), dynamic ensemble selection (DES-ML), Random Forest (RF), and Naive Bayes (NB)—under varying labeling rates to enhance segmentation accuracy. While we acknowledge that deep learning-based methods may offer improved accuracy in some classification tasks, especially with high-dimensional or unstructured data, such approaches typically require large, labeled datasets and introduce higher model complexity. Since expert segmentation labels are only available for a single year and we aim for transparency and interpretability across time, in this study, we employ an unsupervised FCM approach, ensuring consistent and unbiased segmentation across multiple years. To further improve segmentation quality, we first apply Principal Component Analysis (PCA) for dimensionality reduction before implementing the FCM algorithm. By segmenting the entire 2022 Swiss Vehicle Stock Database, we analyze the temporal development of vehicle characteristics across seven main vehicle classes (micro, small, medium, upper-medium, large and luxury, sport, and multi-purpose vehicles), two sub-classes (SUV and non-SUV), and multiple powertrain technologies.

Our findings not only provide insights into the trends in vehicle characteristics at both the class and sub-class levels but also lay the groundwork for more precise fleet modeling. Moreover, they offer valuable insights for fleet decarbonization efforts, infrastructure planning, and the development of effective environmental policies.

The remainder of the paper is structured as follows: Section 2 describes the materials and methods used in this study, including an overview of the datasets (Section 2.1), the segmentation approach (Section 2.2), and its validation (Section 2.3). Section 3 presents the results of our analysis, including the trends in the composition of the Swiss passenger car fleet (Section 3.1), changes in vehicle dimensions and weight (Section 3.2) and developments in engine characteristics (Section 3.3) and the main electric vehicle powertrains (Section 3.4). Finally, Section 4 provides a discussion of the findings, including a summary of the results, links to external influencing factors, the environmental and public health impacts of the observed trends, and the limitations and outlook of the study.

2. Materials and Methods

2.1. Datasets

The following datasets were used to analyze the temporal development of vehicle and engine characteristics in the Swiss passenger car fleet:

• New Registrations (NR) database: This database contains information on all new vehicle registrations in Switzerland at the end of each year. For the validation of our mathematical segmentation approach, we used the NR 2018 dataset, which is available for download from the website of the Swiss Federal Roads Office [33]. The NR 2018 dataset includes 413,431 vehicles and provides details on the vehicle category, make and model, type approval number, drivetrain technology, wheel drive type, number of seats, weight specifications (payload and total vehicle weight), engine capacity, and engine power.
• Vehicle Stock (VS) database: Similar in structure to the NR database, the VS database contains information on all vehicles in circulation throughout Switzerland at the end of the given year. Since annual NR datasets are only available from 2010 onwards, we used the VS 2022 dataset [33], which includes 6,821,551 vehicles, for our temporal analyses. By classifying vehicles based on their first year of registration, we were able to characterize changes in Swiss new vehicle fleets from 1987 to 2022. However, it is important to note that the results for a given year represent only the fraction of registered vehicles that were still in circulation at the end of 2022.
• “Targa” database: The Targa database provides detailed information on vehicle emissions and energy/fuel consumption for each type-approved vehicle model in Switzerland since 1985 and is available for download online [33]. It includes two key datasets: the “Emission” dataset, which contains information on pollutant emissions, and the “Energy Consumption” dataset, which includes vehicle dimensions (length, width and height, reported as ranges when a type approval number corresponds to multiple vehicle variations), fuel type, and energy consumption based on the New European Driving Cycle (NEDC) [34] and Worldwide Harmonized Light Vehicles Test Procedure (WLTP) [35] test cycles.
• “Expert segmentation” database: For validation purposes, we used an expert segmentation database provided by Auto-i-dat [36], which includes segmentation information for approximately two-thirds of the new vehicles registered in Switzerland in 2018.

2.2. Segmentation Approach

In this section, we briefly describe our mathematical approach for segmenting passenger vehicles into seven main classes (micro, small, medium, upper-medium, large and luxury, sport and multi-purpose) and two sub-classes (SUV and non-SUV) based on their dimensions and other technical specifications. Our segmentation method builds upon the machine learning approaches presented in [17,32], incorporating several simplifications and refinements to enhance both the accuracy and applicability across different datasets. Figure 1 provides a schematic overview of the segmentation system model, which is described in detail in the following sections.

2.2.1. Data Pre-Processing

In the first step, we filtered the NR 2018 and VS 2022 datasets to extract passenger cars, exclude direct imports, and check the data for completeness and accuracy. This resulted in 296,041 passenger cars from the NR 2018 and 4,237,494 from the VS 2022 datasets. Less than 2% of the entries were removed due to incomplete information, which is unlikely to impact our findings significantly.

Next, we classified the VS 2022 dataset based on the first registration year (see Figure 2). We then grouped the entries within each year by type approval number and transmission type as different transmission types (automatic vs. manual) affect emissions. The resulting unique vehicle samples for each year were then merged with relevant parameters from the Targa datasets. It is important to note that for the segmentation procedure, we used the grouped unique vehicle samples, while for the segment-based temporal analyses of the new passenger car fleet, we considered all entries in the VS dataset after the initial filtering.

2.2.2. Sports and Multi-Purpose Vehicle Categorization Based on Logical Conditions

For each year in the VS 2022 dataset and for the NR 2018 dataset, we categorized sport and multi-purpose vehicles based on their technical characteristics using the following conditional sequences:

• {Let:
  • Pmin be the minimum power (in kW),
  • Smax be the maximum number of seats,
  • H be the average height of all vehicles belonging to a certain vehicle model (in mm),
  • 4WD be a boolean indicating if the vehicle has a 4-wheel drive (True or False), where the most common drive type within a vehicle model sample is picked for the evaluation,
• Then:
  • Sport vehicles: (Pmin>70 ∧ Smax = 2) ∨ (Pmin>100 ∧ (Smax = 3 ∨ Smax = 4))
  • Multi-purpose vehicles: (H>1700 ∧ 4WD = False) ∨ (H > 1850 ∧ Smax > 5 ∧ Pmin ≤ 125)}

2.2.3. Segmentation of Micro, Small, Medium, Upper-Medium and Large and Luxury Vehicles Using Machine Learning Techniques

After extracting the sport and multi-purpose vehicles, we segmented the remaining vehicle samples separately for each year in the VS 2022 dataset and for the NR 2018 into five main size categories, following the machine-learning based procedure illustrated in Figure 3. Based on the analysis of the expert segmentation, we identified a strong correlation between vehicle width and length and the five main vehicle classes (see Figure 4a). Consequently, the segmentation was performed using the average vehicle model length and width. To fully capture the dimensional variance of the dataset, we first conducted a PCA for each year. Subsequently, we applied the FCM algorithm to the first principal component (PC1) to determine the cluster memberships for each unique vehicle sample in the given year (see Figure 4b). To enhance the convergence behavior of the algorithm, we used the K-means++ initialization method [37]. Lastly, for each year, the resulting classes were assigned to the five main size classes: micro, small, medium, upper-medium, and large and luxury vehicles. In addition, for the NR 2018 dataset, the resulting clusters were then compared to the expert segmentation for validation purposes, as shown in Figure 4c.

2.2.4. SUV/Non-SUV Categorization Based on Logical Conditions

In [32], we demonstrated the application of Semi-Supervised Fuzzy C-Means with a feature learning technique for the intra-class classification of SUV and non-SUV vehicles. In this study, we developed a simpler approach to make the method more widely applicable and easier to implement. Specifically, we categorized SUVs based on the following conditions:

• {Let:
  • H be the average height of the vehicle (in mm)
  • W be the average width of the vehicle (in mm)
• Then:
  • SUVs: (H > 1550) ∧ (W > 1750)
  • Non-SUVs: ¬((H > 1550) ∧ (W > 1750))}

This intra-class categorization was applied to all vehicle model samples in each year for the VS 2022 dataset and the NR 2018 dataset, except those categorized as sport and multi-purpose vehicles, which, with only very few exceptions, were found to always be non-SUVs. Figure 5 shows the expert SUV categorization for the 2018 dataset along with the height and width limits used for our SUV definition.

Note that we also successfully tested the applicability of FCM for the sub-categorization of SUV vehicles in the NR 2018 dataset. However, this approach failed in cases where one of the sub-classes was highly dominant, such as in earlier years with a low share of SUVs.

2.3. Method Validation

To validate the segmentation procedure, we compared the predicted classes and sub-classes for the NR 2018 dataset with those defined in the expert segmentation. Since the expert database specifies only five main size classes, with sport and multi-purpose vehicles treated as separate sub-categories, our validation focuses only on the five main size classes. The validation results, weighted by the total number of cars in the NR 2018 dataset, are summarized in

Table 1. True Positive Rate (TPR) and Positive Predictive Value (PPV) for the size segmentation (micro, small, lower middle, upper middle, large and luxury vehicles) and the SUV/non-SUV categorization of new passenger cars registered in Switzerland in 2018 (the NR 2018 dataset), evaluated against the expert segmentation.

	TPR (%)	PPV (%)
Size Segmentation	82.4	83.4
Micro	85.9	74.3
Small	90.0	86.4
Lower Middle	80.8	85.8
Upper Middle	82.4	69.3
Large and Luxury	73.3	83.0
SUV/Non-SUV	89.0	80.0

. These results show that both the main segmentation method and the SUV/non-SUV categorization achieve True Positive Rates (TPRs) and Positive Predictive Values (PPVs) ranging between 70% and 90%, indicating a high level of agreement with the expert segmentation.

To evaluate the clustering approach for each year in the VS 2022 dataset, we computed the Davies–Bouldin Index (DBI) and Silhouette score. The DBI, which measures the compactness and separation of the size segments (with lower values indicating better clustering), ranged from 1.98 to 8.77 across the years, with an overall average of 3.82. The yearly average Silhouette scores, which indicate the degree of separation between size segments (with higher values indicating better clustering), ranged from 0.22 to 0.36, with an overall average of 0.28.

While the DBI and Silhouette scores varied over the years, the relatively high DBI values and low Silhouette scores suggest limited separation and clustering performance. However, this result aligns with the nature of the input data—vehicle width and length—which exhibit continuous variation and lack distinct boundaries between size classes, as illustrated in Figure 4a. As a consequence, as shown in Figure 4c, our segmentation method, while generally effective, underperforms for vehicles whose width and length are in the boundaries of two categories. These findings underscore the challenges of segmenting data with overlapping distributions and highlight the continuous nature of vehicle dimensions, where defining clear size class boundaries remains difficult. Additionally, expert segmentation, considered as the ground truth for our validation, is inherently subjective and can vary between experts, further complicating the establishment of a definitive ground truth for segmentation and validation. For example, the second-generation Renault Koleos, with a width of 1.843 m and a length of 4.672 m, is reported as a lower-middle-class vehicle in the expert segmentation. However, in agreement with our segmentation results, it is classified as a D-segment (upper middle class) in various automotive classification sources [38]. Similarly, the Suzuki SX4 S-CROSS, with a width of 1.782 m and a height of 1.580 m, is classified as a limousine (non-SUV) in the expert segmentation, whereas, in agreement with our segmentation results, it is defined as an SUV in other automotive sources [39]. Therefore, the inherent subjectivity of expert segmentation, combined with the continuous nature of vehicle dimensions, highlights the need for the cautious interpretation of validation metrics and suggests that rigid classification boundaries may not fully capture the complexity of real-world vehicle categorization.

3. Results

In this section, we present the evaluation of temporal changes in passenger car characteristics based on the Swiss VS 2022 dataset. It is important to note that the results for each year do not represent the entire fleet of cars registered in that year but only the subsample that remained in circulation until 2022. This limitation could introduce biases, particularly for earlier years, as certain vehicle types may have longer lifespans and thus be overrepresented. Additionally, we included only vehicles registered since 1995, as the number of older vehicles still in circulation is negligible. Finally, to ensure statistical significance and meaningful comparisons, we excluded any class (size class or powertrain class) with fewer than 100 vehicles for a given year from our figures. As a result, hydrogen passenger cars are generally not shown.

3.1. Composition of the Swiss Passenger Car Fleet

Figure 6 and Figure 7 illustrate the evolution of the composition of the Swiss passenger car fleet. While some of the observed fluctuations in specific vehicle categories may partially be influenced by segmentation inaccuracies—particularly in cases where vehicle dimensions lie near category boundaries—the overall temporal trends are expected to stem from actual market developments, including evolving customer preferences and shifts in vehicle design. In particular, Figure 6a shows changes in the share of different vehicle classes over time. Small, lower-middle, and upper-middle vehicles constitute the largest share of the fleet. At first glance, the share of sports cars and large and luxury vehicles appears to decrease significantly between 1995 and 2005. However, this trend is likely due to a bias caused by these vehicles remaining in circulation longer than others, leading to their overrepresentation in older samples. Focusing on the period since 2005, we observe a decline in the share of micro and, to some extent, small vehicles, accompanied by an increase in the share of lower- and upper-middle-class vehicles. The share of large and luxury and sport vehicles, shows a slight decrease, while the proportion of multi-purpose vehicles remains relatively stable. In addition, Figure 6b presents the temporal evolution of the SUV share within different vehicle classes, highlighting a continuous increase in their prevalence since the early 2000s. This trend is particularly pronounced in the small, lower-middle, upper-middle, and large and luxury classes. The most substantial increase is observed in the lower- and upper-middle categories, where more than 15% of the new vehicles were SUVs in 2022.

Figure 7 illustrates the temporal evolution of the share of different powertrain types within the various vehicle classes. Since 1995, the share of gasoline vehicles has steadily decreased across all vehicle classes, while the share of diesel vehicles experienced substantial growth in most segments between 1995 and 2017. However, starting in 2017—and accelerating significantly since 2020—there has been a sharp rise in the adoption of alternative powertrains. This trend is particularly evident in electrified vehicles, including BEVs, gasoline HEVs, and gasoline PHEVs, across the micro, small, lower-middle, upper-middle, and large and luxury classes. In the sport class, BEVs and gasoline HEVs have seen a notable increase since 2012 and 2019, respectively, though gasoline remains the dominant powertrain, accounting for over 75% of sport car registrations in 2022. In the multi-purpose vehicle class, diesel remains the most common powertrain, followed by gasoline, accounting for 58% and 28% of registrations in 2022, respectively. While the share of BEVs has been gradually increasing since 2020, and gasoline PHEVs gained some traction in 2022, the overall transition to alternative powertrains in this category has been slower compared to other vehicle classes.

3.2. Vehicle Dimensions and Weight

In this section, we analyze the temporal changes in vehicle dimensions and weight and the corresponding differences among vehicle classes, SUV subclasses, and powertrains.

Figure 8 illustrates the evolution of vehicle width, length, height, and weight in the Swiss passenger car fleet from 1995 to 2022, revealing a clear trend toward larger and heavier vehicles. In particular, all vehicle classes exhibit a steady increase in width, length, and weight, while overall height shows only minor growth despite the rising prevalence of SUVs. Notably, in all cases, the width, length, and weight of a given vehicle class in 2022 are very close to—or have already surpassed—the corresponding dimensions and weight of the next size class in 1995. This pattern, observed across the micro, small, lower-middle, upper-middle, and large and luxury vehicle classes, underscores a significant shift in vehicle design over the past 20 to 30 years.

Table 2. Average vehicle dimensions and weight for main passenger vehicle classes in selected years (2000, 2010, 2020), along with percentage changes since 2000.

Class	Year	Width		Length		Height		Weight
		Mm	Change	Mm	Change	Mm	Change	kg	Change
Micro	2000	1558		3089		1504		1155
	2010	1624	5.7%	3470	4.1%	1518	−1.3%	1309	9.1%
	2020	1651	7.4%	3620	8.6%	1537	−0.1%	1394	16.1%
Small	2000	1654		3753		1487		1446
	2010	1708	2.9%	3960	4.6%	1501	4.1%	1582	9.1%
	2020	1747	5.3%	4057	7.2%	1509	4.6%	1697	17.0%
Lower-middle	2000	1721		4188		1470		1749
	2010	1772	2.1%	4304	3.7%	1536	6.7%	1876	9.1%
	2020	1817	4.7%	4408	6.2%	1545	7.3%	2015	17.2%
Upper-middle	2000	1738		4482		1494		1959
	2010	1811	4.4%	4542	1.2%	1572	10.1%	2106	11.1%
	2020	1857	7.0%	4676	4.2%	1578	10.5%	2302	21.5%
Large and Luxury	2000	1799		4787		1507		2197
	2010	1855	3.1%	4847	1.5%	1547	5.7%	2355	10.5%
	2020	1938	7.7%	4954	3.7%	1640	12.1%	2777	30.4%
Sport	2000	1749		4221		1304		1652
	2010	1832	4.7%	4527	11.8%	1373	6.2%	2014	23.9%
	2020	1850	5.7%	4373	8.0%	1373	6.2%	1909	17.5%
MPV	2000	1821		4640		1917		2420
	2010	1861	2.1%	4752	1.2%	1889	3.4%	2533	3.4%
	2020	1892	3.8%	4899	4.3%	1877	2.7%	2689	9.8%

presents the percentage increases in vehicle dimensions since 2000. Between 2000 and 2020, vehicle width increased across all categories, ranging from 3.8% for multi-purpose vehicles to 7.7% for large and luxury cars. Length also grew, with increases ranging from 3.7% in the large and luxury segment to 8.6% in the micro car category. The most substantial growth is observed in vehicle weight, rising by 9.8% for multi-purpose vehicles and up to 30.4% for large and luxury cars. Changes in vehicle height are less uniform; while micro cars show a slight decrease of 0.1%, other categories experienced increases between 2.7% (multi-purpose vehicles) and 12.1% (large and luxury vehicles).

Figure 9 illustrates the variability in vehicle dimensions and weight across different classes and subclasses in 2022, including all vehicles combined, non-SUVs, SUVs, and various powertrain configurations. Hydrogen vehicles are omitted due to their low registration numbers, which lack statistical significance. The analysis highlights that SUVs generally have larger dimensions—especially in height—and greater weight compared to non-SUVs. Notable differences also emerge across powertrains: gasoline, gasoline HEVs, and compressed natural gas (CNG) vehicles tend to be smaller and lighter than the average. BEVs, while of average size, are heavier due to the battery weight. Diesel and diesel HEVs are typically larger and heavier than their gasoline counterparts, exceeding the overall average. In addition, gasoline and diesel PHEVs tend to be larger and heavier than conventional gasoline, diesel, and hybrid vehicles. Gasoline PHEVs exhibit dimensions and weight similar to BEVs, whereas diesel PHEVs are even larger and heavier. This reflects the characteristics of diesel engines, their prevalence in larger vehicle segments, and the added weight of hybrid and plug-in hybrid systems. While diesel PHEVs appear to be the largest and heaviest vehicles in the comparison, their market share remains minimal (as shown in Figure 7), with no significant growth expected in the future.

3.3. Engine Characteristics

In this section, we analyze the temporal changes in engine characteristics and their variations across different vehicle classes, subclasses, and powertrains.

Figure 10 illustrates the evolution of engine power, energy consumption and tailpipe CO₂ emissions in Swiss passenger vehicles from 1995 to 2022 across the main vehicle classes. The results highlight a general trend toward more powerful and efficient vehicles, with reductions in energy consumption and emissions. A gradual increase in engine power is observed across all vehicle classes, particularly in upper-middle, large and luxury, and sports vehicles from around 2015 onward. Conversely, energy consumption and CO₂ emissions declined across all classes until approximately 2016. These trends reflect significant technological advancements in powertrain efficiency over recent decades, which have led to substantial reductions in energy consumption and emissions, despite simultaneous increases in engine power and vehicle weight. However, since 2017, energy consumption and emissions have shown a slight upward trend. This can be attributed to the sharp rise in engine power and vehicle weight, which technological advancements have been unable to fully compensate for, as well as the transition from the NEDC to the more realistic WLTP test cycle for type-approval certification in 2021. More recently, this increase has slowed or even reversed in some cases, driven by the growing share of PHEVs and BEVs across all vehicle classes.

Table 3. Average engine power, energy consumption and normative CO₂ emissions for main passenger vehicle classes in selected years (2000, 2010, 2020), along with percentage changes since 2000.

Class	Year	Engine Power		Energy Consumption		CO2 Emissions
		kW	Change	kWh/100 km	Change	g/km	Change
Micro	2000	44		50		139
	2010	55	37.3%	46	−14.8%	124	−16.2%
	2020	61	49.8%	36	−33.7%	85	−42.5%
Small	2000	62		59		163
	2010	76	26.1%	51	−10.6%	138	−13.5%
	2020	89	48.4%	40	−30.2%	100	−37.7%
Lower-middle	2000	91		72		198
	2010	104	21.8%	59	−18.3%	159	−19.8%
	2020	126	48.5%	47	−34.0%	123	−37.6%
Upper-middle	2000	120		84		232
	2010	117	6.1%	64	−22.4%	172	−24.5%
	2020	176	59.7%	50	−39.7%	127	−44.2%
Large and Luxury	2000	147		94		259
	2010	156	10.0%	72	−22.1%	195	−23.4%
	2020	245	72.6%	60	−35.0%	155	−39.1%
Sport	2000	154		88		243
	2010	205	44.2%	78	−6.1%	211	−8.1%
	2020	281	97.7%	69	−17.0%	174	−24.4%
MPV	2000	89		86		236
	2010	99	16.0%	72	−14.5%	193	−16.6%
	2020	111	30.5%	61	−27.0%	163	−29.9%

presents the percentage changes in engine power, energy consumption, and CO₂ emissions since 2000 for the main vehicle classes. Between 2000 and 2020, engine power has increased significantly across all vehicle classes, ranging from 30.5% for multi-purpose vehicles to 97.7% for sport cars. Micro, small, lower-middle, sport and multi-purpose vehicles exhibit a consistent upward trend in energy consumption throughout the analyzed period. In contrast, energy consumption in the upper-middle and large and luxury classes primarily increased between 2010 and 2020. Despite these increases, the median energy consumption and normative CO₂ emissions steadily decline across all vehicle classes from 2000 to 2020. Energy consumption reductions range between 17.0% for sports cars to 39.7% for upper-middle cars, leading to a corresponding decrease in normative CO₂ emissions between 24.4% for sports cars and 44.2% for upper-middle cars.

Figure 11 illustrates the variability in engine power, energy consumption, and normative tailpipe CO₂ emissions across various vehicle classes and subclasses in 2022, including all vehicles, non-SUVs, SUVs, and different powertrain configurations. As with previous analyses, hydrogen vehicles were excluded due to their low statistical significance. The results show that SUVs and electrified vehicles generally have more powerful engines than non-SUVs and their non-electrified counterparts. As expected, CNG vehicles, primarily found in the small and lower-middle car segments, as shown in Figure 7, exhibit the lowest engine power. In contrast, BEVs have the highest median engine power and the greatest variability, reflecting their presence across all size classes and their significant share in the upper-middle and large and luxury segments in 2022. Regarding energy consumption and normative tailpipe CO₂ emissions, no major differences are observed between SUVs and non-SUVs. However, clear distinctions emerge among different powertrains. As expected, PHEVs and BEVs demonstrate significant advantages, with energy consumption reduced to approximately one-third and emissions to one-quarter (PHEVs) or even zero (BEVs) compared to conventional powertrains.

Our analysis of all passenger cars registered in Switzerland in 2022, based on the VS 2022 dataset, yields an average CO₂ emission of 124.7 g/km. This closely aligns with the official CO₂ emissions for new car registrations in Switzerland in 2022, reported at 120.9 g/km by the SFOE [40]. The slight discrepancy is likely due to our data pre-processing—such as the exclusion of incomplete records and direct imports—and potential variations between the NR 2022 dataset and the 2022 data extracted from the VS dataset.

3.4. Electric Vehicles

In this section, we analyze the temporal evolution of energy consumption and electric range for the main electric vehicle powertrains.

Figure 12 illustrates the trends in energy consumption and electric range for BEVs and gasoline PHEVs across the main vehicle classes from 2010 to 2022. Only years and vehicle classes with a statistically relevant number of vehicles (>100) are included. For BEVs, we observe a consistent increase in energy consumption across all vehicle classes. This trend reflects the growing vehicle weight and engine power, which technological advancements can no longer fully offset. In contrast, gasoline PHEVs—primarily found in the lower-middle, upper-middle, and large and luxury classes—show a relatively stable energy consumption over time. Regarding electric range, both powertrains exhibit a general upward trend. However, the transition from the NEDC to the WLTP test cycle in 2021 results in an apparent decrease in range for some cases.

Table 4. Average energy consumption (EC) and electric range (ER) of battery electric vehicles (BEVs) and gasoline plug-in hybrid electric vehicles (G-PHEVs) for main passenger vehicle classes in selected years (2015, 2020), along with percentage changes since 2015.

Class	Year	EC BEV		EC G-PHEV		ER BEV		ER G-PHEV
		kWh/ 100 km	Change	kWh/ 100 km	Change	km	Change	km	Change
Micro	2015	13.3		--		152		--
	2020	14.3	7.7%	--	--	222	45.8%	--	--
Small	2015	14.4		--		218		--
	2020	18.0	24.7%	14.6	--	350	60.4%	48	--
Lower-middle	2015	15.5		14.4		195		47
	2020	18.7	20.7%	15.3	6.2%	328	68.1%	55	18.0%
Upper-middle	2015	--		16.6		--		50
	2020	16.7	--	16.0	−3.7%	528	--	56	11.4%
Large and Luxury	2015	22.1		29.2		478		35
	2020	23.2	5.2%	20.0	−31.5%	469	−1.8%	62	77.0%
Sport	2015	12.1		21.8		167		36
	2020	14.2	17.8%	19.8	−9.3%	326	94.6%	54	50.7%
MPV	2015	16.5		--		170		--
	2020	25.5	54.4%	--	--	243	42.7%	--	--

presents the percentage changes in energy consumption and electric range for BEV and gasoline PHEV passenger cars between 2015 and 2020 across the main vehicle classes with statistically relevant data. For BEVs, engine power increases notably across all relevant vehicle classes, ranging from 5.2% in the large and luxury class to 54.4% in multi-purpose vehicles. In contrast, changes in energy consumption for gasoline PHEVs are more variable, with a reduction of 31.5% in large and luxury cars and an increase of 6.2% in lower-middle cars. The electric range of BEVs shows a significant increase between 42.7 and 94.6% across all vehicle classes, except for large and luxury cars, which show a slight decrease of 1.8%. For gasoline PHEVs, the electric range increases between 11.4% in upper-middle cars and 77.0% in large and luxury cars.

4. Discussion

The application of machine learning techniques to segment passenger cars has, for the first time, enabled a detailed class- and subclass-based analysis of the temporal evolution of key characteristics within the Swiss passenger car fleet. Our study goes beyond previous research by not only segmenting the fleet but also extending the analysis beyond 2016, capturing significant developments, particularly the growing relevance of electric powertrains.

Our findings reveal several key trends that illustrate the evolving landscape of vehicle characteristics in Switzerland. First, the share of micro and small vehicles has declined, while lower- and upper-middle-class vehicles and SUVs have gained prominence. Regarding powertrain technologies, we observe a significant increase in the adoption of alternative powertrains, particularly BEVs, gasoline HEVs, and PHEVs, across multiple vehicle classes. This growth has been especially strong since 2020, likely driven by policy measures promoting electric vehicles and the expansion of public charging infrastructure.

Another notable trend is the shift towards larger and heavier vehicles, with significant increases in width, length, and weight since 1995, particularly among micro, small, lower-middle, upper-middle, and large and luxury cars. Unlike previous studies, which suggested a stabilization in vehicle size and weight between 2006 and 2015, our analysis shows no such plateau. Instead, the Swiss new passenger car fleet has continued to grow beyond this period. This trend can be attributed to consumer preferences for more spacious vehicles with increased baggage and cabin space, the additional room required for modern safety features (e.g., crumple zones), and the automotive industry’s interest in marketing larger, more expensive models. Looking ahead, while vehicle width may be constrained by urban road dimensions, vehicle height could continue to increase due to the ongoing popularity of SUVs.

Our analysis also shows that vehicles have become more powerful and efficient over the past few decades, reflecting both technological advancements and shifting consumer preferences. Engine power has increased significantly, particularly since 2015, among upper-middle, large and luxury, and sports vehicles. However, while energy consumption and CO₂ emissions declined until around 2016—consistent with previous studies—our findings indicate a slight increase since 2017. This shift is likely driven by rising engine power and vehicle weight and the transition from the NEDC to the more realistic WLTP test cycle for type-approval certification.

Our findings underscore the complex interplay between technological advancements, consumer preferences, and policy interventions, emphasizing both the challenges and opportunities in guiding the vehicle market toward more sustainable choices. The continued growth in vehicle size and weight presents significant environmental and public health concerns. Larger and heavier vehicles generally consume more energy and emit more CO₂, contributing to climate change and deteriorating air quality. In addition, increased vehicle weight leads to higher emissions from tire and brake wear, which negatively impact air quality and public health. Thus, despite the growing adoption of alternative powertrains, the ongoing trend toward larger, heavier, and more powerful vehicles remains a major challenge in reducing the environmental footprint of the passenger car fleet.

Beyond environmental concerns, these trends also pose challenges for urban infrastructure. As vehicles grow in size, existing parking spaces, road widths, and traffic management systems may require significant redesigns or expansions to accommodate larger vehicles. The increasing number of electric vehicles further necessitates a substantial expansion and upgrade of charging infrastructure, especially in densely populated urban areas where access to private charging facilities is limited. In Switzerland, this could mean rethinking the design of residential and commercial parking areas and increasing the availability of public charging stations. Furthermore, there may be a need for policy incentives to encourage the construction of infrastructure that supports both larger vehicles and the growing fleet of electric vehicles. The increasing electrification of the vehicle stock also introduces new challenges related to grid capacity, making it essential to align the expansion of charging infrastructure with energy demand and supply. These considerations are not unique to Switzerland and are equally relevant for other countries facing similar trends in urbanization and vehicle characteristics.

While our methodologies have been developed, validated, and applied to the Swiss passenger car fleet, they can be readily adapted to analyze other fleets and countries. The workflow is flexible and primarily data-driven, which enables its transferability to other national contexts, provided that similar vehicle registration and technical datasets are available. However, our analysis is based on type-approval values, and while the WLTP cycle provides a more realistic assessment than the NEDC test cycle, it still does not account for many real-world factors influencing energy consumption and CO₂ emissions, such as auxiliary energy use, real-world driving profiles, and actual utility factors and charging profiles for PHEVs and BEVs. Future research should therefore focus on linking these analyses to real-world operational data to provide a more accurate understanding of vehicle performance and environmental impact. Acquiring detailed vehicle and operational data remains a challenge due to the complexity and resource requirements involved, but overcoming this barrier is crucial for bridging the gap between type-approval and real-world performance. A more comprehensive understanding of the environmental impacts and efficiency gains across different vehicle classes and powertrains will ultimately support better policy decisions and guide consumers toward more sustainable transportation choices.

Although the current model provides a detailed analysis of past vehicle characteristics and trends, it could offer valuable insights for predicting future trends. Several factors are likely to influence these trends, including socio-demographic shifts (such as urbanization, income levels, and lifestyle preferences), technological innovations (e.g., electric vehicles, hydrogen fuel cells, and autonomous driving), policy interventions (such as environmental regulations and government incentives), and changes in consumer preferences and mobility trends (e.g., the rise in shared mobility options). While incorporating these elements into the model would require extending its scope with additional predictive components and data sources, this remains a promising direction for future research, offering the potential to better forecast fleet evolution and guide strategic decision making.