Public transport is critical in facilitating the functioning of cities (Hernandez, 2018; Kalaoane et al., 2024). The need for quality, effective, and efficient public transport systems gained traction toward the turn of the 21st century, thus the adoption and popularity of the sustainable public transport discourse that made its way into urban strategies and policies in African cities (Sam et al., 2018; Oqubay, 2025). Ochoa-Covarrubias et al. (2021) state that sustainable transport provision refers to “access to safe, affordable, accessible and sustainable transport systems for all, improving road safety, notably by expanding public transport, with special attention to the needs of those in vulnerable situations. The Sustainable Development Goals (SDGs) provide a global framework for addressing various challenges facing cities, including inequality, poverty, and climate change (Tazzie et al., 2024). Transport classification often aids SDGs by monitoring and enhancing transportation systems, particularly Goal 11, which involves promoting sustainable cities and communities (Almulhim et al., 2024). Furthermore, the classification of transportation modes facilitates the creation of an intelligent, robust transportation infrastructure. Additionally, classification aids in climate monitoring and mitigation strategies that address Goal 13 on Climate Change (Owusu-Sekyere et al., 2024).

Nevertheless, the state of public transport services and infrastructure is seemingly declining across the African continent (Yankson, 2022), thus affecting people who depend solely on public transport to take part in economic life (Borker, 2022). The poor state of public transport systems undermines the strive for sustainable cities and communities. Díez-Mesa et al. (2018) and Stokenberga et al. (2025) acknowledge that existing mass transit is unable to give full user satisfaction. Consequently, in Johannesburg, the quality of transport services in Johannesburg is marked by both efforts at modernization and persistent systemic challenges. While services like Metrobus and the Rea Vaya Bus Rapid Transit (BRT) system have improved access and reduced carbon emissions (Matubatuba and De Meyer-Heydenrych, 2022; Almassawa et al., 2025; Yildirim and Akin, 2025), more efficient vehicles, users still face significant issues such as long travel times, inconsistent service, and overcrowding. According to the 2019/20 Gauteng Household Travel Survey (GHTS), nearly 60% of households in Gauteng spent more than the recommended 10% of their income on public transport, up from 55% in 2014. This trend is particularly burdensome for lower-income households, who are disproportionately affected by rising transport costs. The survey highlights that transport continues to contribute significantly to the increased cost of living in Gauteng, with poorer households being the most impacted. These findings underscore the need for improved and affordable public transport options given the African context.

In this study, the objective of the classification model is to find the main attributes that distinguish two modes of transport from each other based on users’ evaluation of their performance. Ultimately, establish a user-centric transportation system that is sustainable, inclusive, adaptive, and data-driven for the creation of sustainable cities. The paper is structured as follows: Section 1 presents an introduction, and Section 2 discusses related works and contributions. In Section 3, methodological consideration is presented, including data acquisition process, in addition to the methodology for developing the classification algorithm, including the selection of relevant features and the implementation of SVM. In Section 4, the results obtained from evaluating the performance of the SVM algorithm are presented. Discussions and implications of the findings are discussed, and potential future directions are explored in Sections 5 and 6, respectively. Finally, conclusions are presented in Section 7, and limitations of the study are discussed in Section 8.

The application of Machine Learning (ML) to transportation systems has been increasingly popular in recent years. ML models are becoming essential in transport analytics, from demand forecasting (Demissie et al., 2016; Profillidis and Botzoris, 2018; Nguyen et al., 2020; Ghorbani et al., 2025) and mode detection to traffic prediction and route optimization (Nagy and Simon, 2018; Boukerche and Wang, 2020; Liu et al., 2025; Priya and Francis, 2025). Subsequently, models such as SVM, Random Forests, Decision Trees, K-Nearest Neighbor (K-NN), and Neural Networks have been used in transportation mode classification (Kabiri et al., 2023). The intricacy, data needs, interpretability, and generalizability of these models vary. For example, Fang et al. (2016) used SVM, Decision Trees, and K-Nearest Neighbor algorithms to classify transportation and vehicle modes using accelerometer, magnetometer, and gyroscope data from smartphones. SVM demonstrated the best accuracy compared to other ML models. In the context of developing countries, Tamirat et al. (2023) utilized SVM, which proved to be suited for complex rural–urban landscape change and pattern analysis.

In most of these studies, SVM demonstrates the most promising among these in the classification of transport modes, Roy et al. (2020) particularly in cases when the data is nonlinear, sparse, or derived from multiple sources. Kristiyanti et al. (2022) applied SVM in a study that collected data on Light Rail Transit and Mass Rapid Transit from tweets. SVM in this study achieved high accuracy and provided insights into public opinion, guiding improvement in service within the transportation mode. Similarly, AlKhereibi et al. (2022) utilized an SVM-based approach to urban and transport planners in forecasting transit demand and planning land use around metro stations, supporting Transit-Oriented Development (TOD) strategies in Qatar. SVM offers a comparative advantage over other classification algorithms in this study on the efficacy of SVM in categorizing public transportation modes.

Recent studies, such as Hagenauer and Helbich (2017) and Ghorbani et al. (2025), have demonstrated that classifiers like Random Forest, XGBoost, and CatBoost, frequently outperform other classification models across various domains. Although comparative evaluation across multiple algorithms is a valuable approach, the present study focused specifically on assessing the performance of Support Vector Machines (SVM) within the context of classification of different modes of public transport. SVM was ultimately selected based on both theoretical and empirical considerations. SVM is particularly well-suited for high-dimensional spaces and demonstrates strong generalization capabilities. In this study, SVM outperformed other models during the validation phase, achieving a high accuracy of 90%. This performance, combined with SVM’s robustness to overfitting and its effectiveness in handling non-linear decision boundaries, justified its selection as the optimal classifier for this study. This holistic approach allows a better understanding of the consumer’s perspective on the current public transport system.

A structured survey was conducted with 300 commuters from March to April 2023 using a cluster sampling technique. Cluster sampling divides the population of interest into groups and allows random selection from each cluster to represent the sample of the study. This can be done in two ways; single-stage (Wu et al., 2020) and multi-stage cluster sampling (Fawzy et al., 2021). This study followed a multi-stage sampling technique, which did not require a sampling frame (Sharma, 2017). When a thorough sampling frame is lacking or challenging to establish, multi-stage sampling is a useful strategy because it allows a researcher to utilize locations or other groupings as the primary sampling units (Brown, 2010). Data was collected in Braamfontein, an inner-city educational and commercial precinct in the city of Johannesburg. In this study, all public transport stations and stops in Braamfontein were clustered, and participants were selected randomly from those clusters. According to Taherdoost et al. (2022), a sample size of 100 or greater is sufficient for SVM because the greatest significance lies in the data’s ability to generate impactful and robust results.

Consequently, this study utilized data from 300 surveys to ensure robustness. The purpose of the survey instrument was to collect detailed information from users regarding the quality of different modes of transport services. 300 surveys collected in this study reduced the risk of overfitting and improved the robustness and accuracy of the model, ensuring that the results were more representative. The survey consisted of the use of a Likert scale requesting respondents to gauge their satisfaction level from 1 (dissatisfied) to 10 (extremely satisfied) for the level of services provided by different modes of public transport systems (see Table 1), and consent was sought from commuters before probing. In addition, Ethics approval was obtained from the Higher Education Ethics Committee to ensure the study adhered to ethical research standards. The survey was conducted with informed consent from all participants. Data were then analyzed using Support Vector Machine (SVM), a supervised machine learning method that can be applied to regression and classification (Jain et al., 2020).

Developing a well-functional, sustainable, and inclusive transportation system firstly requires to determining the key characteristics that set apart various public transportation options. Identifying a classify to distinguish between modes of transportation enables a detailed examination that goes beyond superficial characteristics, resulting in better system optimization and more informed decision-making. SVM was employed because it effectively classifies modes of public transport based on rated attributes by the users, with a strong accuracy rate.

The classification model serves as a valuable tool for identifying the key attributes that distinguish between two modes of transport; minibus taxis, buses and ride-hailing based on users’ evaluations of their performance. By highlighting the most influential factors that drive user preferences, the model provides evidence-based guidance for transport policy and planning. The primary attributes differentiating minibus taxis and ride-hailing are comfort, affordability, politeness, access to stops, driving styles, speed, and trip duration thus policymakers need to prioritize interventions, and investment should go toward the defining attributes. Upgrading minibus taxis and improving station environments can enhance its effectiveness. Moreover, another key attribute is driver behavior in determining the overall quality of services. Concerns regarding risky driving practices highlights the necessity of focused driver behavior training programs to advance safety on the roads. Although the study acknowledges South Africa’s attempts to transition certain minibus taxis to electric cars, it also points out that driver behavior and the underlying infrastructure have not yet kept pace with these developments including other factors. Lessons can be learnt to enhance the performance and integration of other modes of transportation by analyzing what functions well in one.

Support Vector Machines are a strong and dependable technique for categorizing public transport modes based on travel behavior data, as this study has presented. Each mode of transport examined in the study presents unique strengths, however, the overarching objective is to enable policymakers to identify best practices across modes. By understanding what works well in one transport mode, lessons can be drawn and adapted to improve the performance and integration of others to respond better to consumer needs while maintaining the goal for sustainable transportation.

In this study, the sample size and study area may not have captured a sufficiently large or diverse sample of users, which could impact the generalizability of the findings. Additionally, the reliance on self-reported data introduces potential bias, as respondents may not always accurately reflect their experiences or perceptions. For example, users might overestimate their satisfaction or reduce negative aspects of the service.