Data-Driven Insights into Road Accident Severity in Sub-Saharan Africa: A Multiple Correspondence Analysis for SDG-Aligned Policy

Road traffic accidents continue to pose a serious challenge, especially across Sub-Saharan Africa, which, quite disturbingly, recorded the highest mortality rate globally in 2021, with 19 deaths per 100,000 people [1]. While numerous efforts have been made to identify contributing factors and curb the trend, the statistics remain sobering: road accidents still top the list as the leading cause of death among children and young people aged 5 to 29, and sit as the 12th leading cause of death across all age groups. Beyond the tragic loss of lives and the burden of long-term injuries, the economic impact is also significant, with some countries losing up to 6% of their GDP to costs associated with road crashes [2]. Road traffic accidents, therefore, not only constitute a major health challenge but also a development challenge, requiring equal or more attention to those given to other sustainable development objectives, such as poverty reduction [3, 4].

Much research has investigated factors affecting road accident severity, especially in developing African countries. These include human factors, such as driver inattention, speeding, alcohol consumption, young inexperienced drivers, and the failure to maintain vehicles [5]. Both studies [6, 7] identified speeding as a significant predictor of accident severity, while alcohol consumption increases crash likelihood. Also, behaviors such as dangerous overtaking pedestrian actions, like jaywalking, can have a pronounced impact on accident severity [8]. Another notable factor is vehicle conditions, including vehicle type, age, and maintenance levels. Ouni and Chaibi in reference [8] suggest maintenance conditions as a key contributor to accident outcomes. Environmental factors such as roadway geometry, lighting conditions, and weather were not exempted. For example, poor road infrastructure, such as poor lighting and road designs, contributes to severe accidents [5, 9].

Most of the research has focused on the use of machine learning models, such as decision trees and support vector machines, to identify how these factors affect road accident severity. For instance, the J48 pruned tree model has been shown to outperform other models presented in reference [6]. Recent reviews [10, 11] of several statistical methods applied in road accident analysis show that tangible milestones have been achieved in identifying the factors influencing road traffic accidents and their severity. For example, logistic regression was used to identify risk factors in single-vehicle traffic accidents and to investigate how various factors, like road environmental conditions, affect accident severity [12, 13]. Ordered probit models were used to determine the ordered severity levels of injuries in accidents occurring in a specific area or during a defined period, such as cold and snowy [14]. Among others, a Bayesian Hierarchical Model, an advanced statistical technique, has been applied to assess the role of multi-level data hierarchies in road safety, such as the influence of individual, road, and environmental factors on motorcycle crashes at intersections [15].

While it is widely acknowledged that identifying the specific factors influencing traffic accident severity is important, far less attention has been given to exploring the deeper, often hidden relationships between these variables. These often-overlooked connections, frequently buried beneath layers of categorical complexity, remain underexplored, not least because of the methodological hurdles involved. Yet, understanding these relationships is vital if we are to set meaningful priorities and craft policies that go beyond surface-level fixes. To probe how various factors interplay in shaping accident severity, this study employs Multiple Correspondence Analysis (MCA), a method well-suited to unpacking nuanced patterns that might otherwise go unnoticed. While MCA isn’t a silver bullet, its strength lies in its ability to distill multi-dimensional data into digestible trends, offering a clearer picture of how seemingly disparate variables coalesce. In the context of global road safety goals and the push toward sustainable development, these insights serve a practical purpose: guiding more targeted interventions that can reduce not just the economic toll but also the deeper, often hidden social costs of traffic accidents.

Multiple Correspondence Analysis (MCA), an evolution of Correspondence Analysis (CA), was designed to work specifically with categorical data [33]. At its core, MCA seeks to reveal and map out the hidden relationships within datasets composed of multiple categorical variables. Rather than merely providing tabular summaries, it brings these associations to life through intuitive visualizations, making it easier to spot underlying patterns that might otherwise go unnoticed [34]. In this study, MCA is employed not just for its technical suitability, but also for its ability to expose the geometric configurations of factors influencing road traffic accident severity, something conventional summaries often fail to capture. The process unfolds in three essential stages: preparing the dataset, applying appropriate coding, and performing the analysis.

3.1 Data preparation

The dataset underpinning this study was sourced from manually documented road traffic accident reports spanning 2017 to 2020, yielding a total of 12,316 entries across 32 initially derived features [35]. Since MCA is best suited to categorical data, the dataset comprises a deliberate mix of nominal and ordinal variables. During the pre-processing phase, entries with missing or incomplete information were filtered out, resulting in a more analytically viable subset of 6,422 valid records. To maintain analytical clarity and reduce noise, 22 features out of the original 32 were retained based on their relevance to accident causality. These selected variables, as detailed in Table 1 alongside their category distributions, capture dimensions frequently associated with road traffic outcomes. It should be acknowledged, however, that the dataset is constrained to a specific time frame (2017-2020). While this window provides a solid snapshot of recurring patterns and risk factors, it may not fully reflect the influence of more recent shifts in driving behaviour, safety regulations, or emerging vehicular technologies elements that could reshape the current risk landscape in ways this dataset cannot fully anticipate.

Table 1. Description of road traffic accident features

3.2 Data codification

The MCA categorical data was coded using 6422 records after data pre-processing. The MCA analysis considered features, categories per feature, and frequency of each category. The first 22 features presented in Table 2 are numbered from 1 to 22, coded F1 to F22. Each category within these 22 features is assigned a code ranging from 1 to 159, coded F23-F181. The frequency of each category is classified into four groups, coded 1 to 4, representing very small, small, large, and very large frequencies, respectively coded F182 to F185. The sample size ranges are as follows: 1 to 1600 (very small), 1601 to 3200 (small), 3201 to 4800 (large), and 4801 to 6400 (very large). Table 2 presents the coding for the severity of road accident factors.

Table 2. Severity of road accident factors

3.3 Data analysis

Correspondence Analysis extends the application of CA to more than two categorical variables, generalizing Principal Component Analysis (PCA) for categorical data and revealing patterns in complex datasets [36]. The subset of the dataset focusing on influencing factors and sources of heterogeneity was analyzed using MCA to uncover hidden associations.

MCA presents a methodologically and strategically appropriate choice as the primary analytical technique for this study. This is largely due to the categorical nature of the variables typically encountered in accident severity datasets, such as vehicle type, road condition, time of day, weather, type of collision, and severity level (e.g., fatal, serious, or minor). Unlike traditional statistical techniques that are more suited to numerical or normally distributed data, MCA is specifically designed to accommodate multiple categorical variables simultaneously. This makes it particularly effective in uncovering meaningful patterns and relationships within complex, categorical datasets.

MCA, a type of Correspondence Analysis, can be performed on either an indicator matrix or a Burt matrix, with both matrices being central to the analysis. However, the standard approach to MCA is fundamentally based on the indicator matrix [37]. This matrix presents each observation as a disjunctive map of variables; essentially, each column corresponds to a specific category within a categorical variable [33]. Each observation is assigned a binary indicator: a 1 if it belongs to a specific category and 0 otherwise. To explore this categorical structure, we employed Multiple Correspondence Analysis (MCA) via the MultipleCar toolbox [34], which uses weighted least squares estimation balancing precision with a reasonable degree of robustness in how parameters are derived.

At the heart of this approach lies the disjunctive (indicator) matrix, which reformulates categorical responses into a format amenable to multivariate analysis. While this transformation appears mechanical, it plays a critical role in clarifying hidden associations among variables and lends itself well to visual interpretation, without sacrificing computational simplicity. In practice, the indicator matrix offers more than just structural clarity. It’s relatively easy to construct, keeps the underlying data intact, and brings to light nuances at the individual level. Its flexibility in handling incomplete records is a quiet advantage, especially when working with large, imperfect datasets, something most real-world studies inevitably contend with.

Identifying the relationship between factors that impact accident severity helps make sense of complex datasets to transform raw data into meaningful, actionable insights. Leveraging MCA, the results provide connections among variables and reveal unexpected patterns that might otherwise go unnoticed. The biplot highlights distinct clusters of variables influencing accident severity. While some factors are interconnected, others show more independent influences, providing a comprehensive understanding of the interplay between demographics, environmental conditions, vehicle characteristics, and temporal patterns. With the top-right quadrant showing positive correlations, the results highlight factors such as vehicle ownership, vehicle type, younger driver demographics (18-30), specific lanes or medians, area of occurrence, and weekdays and implies that a cluster of factors such as vehicle ownership, type, and younger driver involvement work together to shape accident severity, revealing opportunities for targeted interventions. The key findings from this study provide valuable insights that can guide stakeholders in identifying critical areas of focus. By aligning their strategies with the study’s findings, stakeholders are better positioned to make thoughtful decisions, focus on what truly matters, and apply solutions where they’ll have the most impact. The points below outline key areas that deserve attention to ensure efforts are effective and outcomes are genuinely meaningful:

1. Government agencies can draw valuable insights from the top-right quadrant to strengthen road safety through well-targeted policies and regulations. For instance, more rigorous licensing and training requirements could be introduced for younger drivers [38], with a focus on improving safe vehicle handling and raising awareness of accident-prone zones [39]. One practical yet often underemphasized approach to improving road safety involves enforcing routine inspections for older vehicles, particularly those owned or operated by younger drivers. While this may sound regulatory, it serves a preventive purpose: ensuring that vehicles on the road remain mechanically sound and less prone to avoidable failures [40]. It might also be time for authorities to revisit existing zoning policies, not merely to review speed limits, but to rethink the logic behind where and how such limits are applied. In areas identified as high-risk zones, this could include layered interventions such as improved signage, speed calming measures, or even infrastructural redesign. Taken together, these actions not only reduce accident frequencies but also contribute to a more proactive and inclusive road safety culture [41].
2. Urban planners can extract actionable insights from the top-right quadrant of the MCA biplot, using them to rethink infrastructure in areas consistently flagged as accident hotspots. Rather than generic fixes, targeted interventions such as clearer lane markings, enhanced lighting around medians, and properly timed pedestrian crossings can significantly improve visibility and lower accident risk. Importantly, the demographic layer, especially trends linked to younger drivers, should not be overlooked. Traffic calming measures in proximity to schools, universities, and recreational centres could prove more effective than blanket enforcement. Moreover, high-risk lane data offers a compelling case for the deployment of intelligent monitoring tools, such as real-time sensors, surveillance cameras, and AI-driven alerts, to provide early warnings and encourage responsible road use before incidents occur [42, 43].
3. Automotive manufacturers are uniquely positioned to contribute meaningfully to road safety, particularly by embedding features tailored to younger drivers, who often fall into high-risk categories. Safety mechanisms like automatic emergency braking, lane departure alerts, and electronic stability control are more than technological upgrades; they are strategic tools for mitigating behavioural risks common in this age group. Beyond engineering, there’s also a marketing gap to address. Messaging that speaks directly to the lifestyle, concerns, and media consumption patterns of young drivers could enhance uptake. There’s certainly room for creativity here: offering customizable safety packages within entry-level models could combine affordability with proactive risk management, making safe driving both attractive and accessible.
4. Healthcare providers, though often seen as reactive stakeholders in the aftermath of road accidents, have a pivotal role to play in reshaping response systems, especially in areas with high accident incidence involving younger drivers. It begins with strengthening trauma systems: well-equipped ambulances, rapid deployment, and skilled first responders can make the critical difference in survival and recovery outcomes. But the opportunity extends beyond emergency response. Collaborations between healthcare institutions and road safety agencies could facilitate life-saving education, basic first aid training, accident response preparedness, and peer-led awareness sessions for young drivers. Additionally, by analysing accident patterns, health professionals are well-placed to advocate for community-level interventions, such as prevention-focused campaigns that stress responsibility over recklessness on the road [44].
5. Traffic authorities can adopt a more dynamic approach to risk reduction by implementing time-sensitive traffic management strategies in zones identified as weekday accident clusters. For example, time-based lane closures or temporary directional shifts could be applied during peak-risk hours, especially on routes heavily used by younger drivers. These measures, though logistically demanding, offer a data-driven way to disrupt patterns of recurring accidents. Equally, the targeted deployment of enforcement personnel based on heatmap insights can enhance compliance and reduce opportunistic traffic violations. This kind of precision in planning not only deters risky behaviour but also builds a visible presence of accountability, which is often lacking in high-incident corridors [45, 46].
6. The Top-Right Quadrant of the MCA biplot offers insurance providers a clear vantage point for refining risk assessment strategies. Premium adjustments that reflect key factors such as vehicle ownership, car type, and driver age can serve both as deterrents and incentives, nudging high-risk groups toward safer practices. However, it’s not just about penalties; insurers could also introduce reward structures—discounts for young drivers completing certified safety programs or investing in advanced vehicle safety features. This dual approach balances caution with encouragement. Moreover, tailored insurance models focused on specific risk profiles allow for proactive engagement rather than reactive claims management. By aligning underwriting with real-world accident patterns, insurers stand to improve both prediction and prevention, all while reinforcing trust through greater transparency.