Risk Assessment and Adaptation Profiling of Non-Standard LPG Installations in Light Commercial Vehicles: Insights from Kumasi, Ghana

Abstract

The rapid rise in the use of Liquefied Petroleum Gas (LPG) as an alternative vehicle fuel in Ghana presents both opportunities and risks within the national energy transition agenda. This study investigates LPG safety as well as environmental and regulatory implications using a multi-method quantitative approach that combines structured survey data, exploratory multivariate analysis (MCA), and machine learning classification (Random Forest) to uncover emerging associations and patterns in LPG safety practices. Primary data were obtained from 384 respondents, including vehicle operators, auto-technicians, regulatory officials, and LPG station attendants across five major transport zones: Kejetia, Asafo, Ahodwo, Bantama, and Suame Magazine. The MCA identified four distinct behavioural and safety profiles—At-Risk, Proactive Safety, Compliant and Equipped, and Formal and Reported—reflecting diverse compliance and risk patterns across socio-occupational groups. The Random Forest classifier achieved a predictive accuracy of 96.5% based on cross-validated performance. Sensitivity and specificity values were high, indicating reliable discrimination among incident types. To reduce the risk of overfitting, k-fold cross-validation and monitored error convergence were performed across increasing numbers of trees. While the model shows strong predictive capability, we present these results cautiously and emphasize observed associations and emerging patterns rather than definitive predictive conclusions. The findings reveal that while economic motivations underpin LPG adoption, weak institutional enforcement and widespread informal installations heighten safety vulnerabilities. Comparisons with sub-Saharan and Asian contexts underscore the need for a structured regulatory framework, mandatory certification of installers, and periodic vehicle inspections. The study contributes to the broader discourse on informal energy transitions in developing economies by demonstrating how technical and behavioural determinants interact within weak regulatory systems. Policy recommendations emphasize the integration of data-driven risk assessment tools into regulatory oversight to enhance vehicular LPG safety and sustainability.

1. Introduction

Emissions from motor vehicles have significant adverse impacts on human health, ecosystems, and the global climate. Road transport remains a major contributor to atmospheric pollution, accounting for substantial shares of carbon dioxide, carbon monoxide, volatile organic compounds, nitrogen oxides, and black carbon emissions worldwide [1,2]. In response, many countries have pursued strategies aimed at reducing vehicular emissions through stricter regulatory standards and the adoption of cleaner alternative fuels [3].

Among alternative transport fuels, liquefied petroleum gas (LPG), natural gas (NG), and hydrogen have received growing attention, with LPG and NG being the most widely adopted in commercial transport due to their relative affordability and established supply chains [4]. LPG—commonly referred to as autogas—is primarily composed of propane and butane and is valued for its cleaner combustion characteristics compared to conventional petrol and diesel fuels [5].

Despite these advantages, the adaptation of LPG for vehicular use in many developing regions is frequently characterized by non-standard and informal conversion practices. In Ghanaian cities, including the Greater Kumasi Metropolis, LPG installations range from professionally executed conversions using certified components to substandard modifications involving uncertified piping, valves, fittings, and, in extreme cases, repurposed domestic cylinders. Such practices are commonly associated with first-generation or rudimentary conversion systems and are prevalent in contexts with limited regulatory oversight [6].

While LPG adoption is often driven by economic incentives and fuel availability, the safety implications of informal conversions raise significant concerns. Poor installation quality and inadequate maintenance increase the likelihood of gas leaks, fire outbreaks, and system failures, undermining both public safety and the environmental benefits of LPG use. Although extensive studies from Europe, North America, and parts of Asia document the performance, emission characteristics, and economic benefits of LPG vehicles [3,4,7], empirical evidence from sub-Saharan Africa remains limited.

In Ghana, existing research has focused primarily on major urban centres such as Accra and Tema, leaving substantial gaps in understanding the prevalence, safety performance, and regulatory responses associated with non-standard LPG vehicle installations in other cities. Kumasi, given its dense commercial transport activity, diverse vehicle fleet, and distinctive enforcement dynamics, warrants focused investigation. A detailed assessment of local installation practices, user awareness, incident histories, and institutional capacity is essential for developing evidence-based interventions that support the safe and sustainable adoption of LPG in the transport sector [5].

Accordingly, this study assesses the adaptation practices and safety implications of non-standard LPG installations in light commercial vehicles within the Greater Kumasi Metropolis of Ghana. Specifically, the study (i) identifies key adaptation needs for improving safety and reducing accident risk, (ii) provides empirical insights based on field observations and stakeholder engagement, and (iii) proposes policy, enforcement, and technical measures to enhance LPG vehicle safety (Appendix A).

The remainder of this paper is organized as follows: Section 2 reviews relevant literature on LPG vehicle conversion practices locally and globally. Section 3 outlines the methodological framework adopted in the study. Section 4 presents and analyses the empirical findings from Kumasi, while Section 5 discusses their implications for safety, regulation, and environmental performance. Section 6 concludes with key findings, limitations, and recommendations for policy and future research.

2. Literature Review

2.1. LPG Adaptation and Safety Practices in Ghana

In Ghana, liquefied petroleum gas (LPG) has historically been promoted as a cleaner alternative for domestic energy use. In recent years, however, its application as an automotive fuel has increased, particularly within urban transport systems. National-level analyses highlight both opportunities and constraints to this transition, including supply chain limitations, uneven refuelling infrastructure, and weak enforcement of conversion standards [8].

Empirical studies conducted in cities such as Tema and Ho indicate that most LPG-powered vehicles in Ghana are retrofitted gasoline engines rather than purpose-built Autogas vehicles [9]. While retrofitting enables rapid adoption, it introduces significant safety concerns when conversions are performed outside formal regulatory frameworks. Many installations are carried out by informal mechanics using uncertified components and improvised methods, increasing the likelihood of gas leaks, pressure regulation failures, and fire incidents [10]. Regulatory agencies, including the Driver and Vehicle Licensing Authority (DVLA), face persistent challenges in enforcing technical standards due to limited inspection capacity and inadequate institutional resources.

Infrastructure-related constraints further shape LPG adoption outcomes. Limited access to refuelling stations, long travel distances to filling points, and uneven spatial distribution of facilities restrict safe and efficient LPG use, particularly within dense urban transport corridors [11]. These limitations are especially relevant in Kumasi, where high taxi and minibus activity, growing fuel demand, and concentrated refuelling locations create both opportunities for LPG expansion and heightened safety risks [12,13].

Despite growing interest in LPG-based transport, there remains a lack of comprehensive empirical evidence examining installer capacity, safety performance, regulatory compliance, and user behaviour in Kumasi. For example, Figure 1 show a poorly secured LPG cylinder installed in a vehicle. Addressing these gaps is essential for developing context-specific policies that support safe LPG adaptation while minimizing accident risk.

2.2. Global Practices in LPG Vehicle Conversion and Regulation

Globally, LPG has gained recognition as a transitional transport fuel due to its relatively low emissions, affordability, and established production as a petroleum by-product [14]. European countries such as Italy, Poland, Turkey, and the Netherlands lead global LPG adoption, supported by robust regulatory frameworks, fiscal incentives, and mandatory safety inspections [15].

Within the European Union, LPG vehicle conversions are strictly regulated: only certified workshops are permitted to carry out installations, and converted vehicles must comply with stringent emission and safety standards verified through periodic inspections and real-driving emissions testing [16]. These measures have resulted in significantly lower nitrogen oxide and particulate emissions and improved operational safety compared to informally retrofitted systems.

In contrast, countries with weak institutional and regulatory frameworks often experience widespread informal conversions and elevated safety risks. Studies across Sub-Saharan Africa indicate that uncertified installations dominate due to cost pressures, limited access to formal training programs, and the absence of enforceable technical guidelines [17,18]. Technical risks associated with such practices include gas leakage, improper combustion, and increased fire hazards, particularly when uncertified materials or poorly secured cylinders are used as shown in Figure 2 [19].

Severe accidents, including Boiling Liquid Expanding Vapour Explosions (BLEVE), have been widely documented in association with non-standard LPG installations [20]. Beyond acute accident risks, prolonged exposure to LPG also poses occupational health hazards for drivers, technicians, and refuelling attendants, ranging from respiratory distress to frostbite injuries during refilling operations [21]. Environmental concerns persist when unburnt LPG is released through leaks or incomplete combustion, contributing to urban air pollution and greenhouse gas emissions.

Overall, international experience demonstrates that the economic and environmental benefits of LPG can only be fully realized under conditions of strict safety regulation, certified installation practices, routine inspections, and sustained public awareness initiatives. These lessons provide a critical benchmark for evaluating LPG adaptation practices in cities such as Kumasi.

3. Materials and Methods

3.1. Study Area

Kumasi, the capital city of the Ashanti Region, is centrally located in Ghana and comprises 43 local government assemblies, including one metropolis, 18 municipalities, and 24 districts (MMDAs). Geographically, the region lies between longitudes 0°9′0″ W and 2°15′0″ W, and latitudes 5°30′0″ N and 7°27′36″ N. According to the 2021 Population and Housing Census, the Ashanti Region has a population of 5,432,485—representing approximately 10.2% of Ghana’s total land area of 24,389 km². Kumasi’s central location enhances its strategic importance for national trade, logistics, and transportation networks.

As one of Ghana’s busiest urban centres, Kumasi serves as a commercial and transport hub characterized by a high density of passenger cars and light commercial vehicles. The city was selected as the study area due to its extensive use of Liquefied Petroleum Gas (LPG) as an alternative fuel for vehicles, driven by the rising cost of petrol and diesel. The transportation sector in Kumasi is dominated by taxi drivers, trotro operators, and private car owners, many of whom resort to non-standard LPG conversions.

Five key areas were purposively selected for field data collection (Kejetia, Asafo, Ahodwo, Bantama, and Suame Magazine) based on their significance in LPG use and conversion practices (Figure 3). Suame Magazine, the largest automobile repair enclave in West Africa, hosts numerous informal auto-technicians specializing in non-standard LPG installations. Kejetia and Asafo were selected for their concentration of commercial transport operators, while Ahodwo and Bantama represented mixed-use zones with both middle-income private car owners and commercial drivers. This stratified selection ensured diverse socio-economic and occupational representation across the sampled population.

3.2. Data Sources, Population, and Sampling

Both primary and secondary data were utilized in this study. The target population comprised vehicle operators (taxi drivers, trotro operators, and private car users), auto-technicians involved in LPG installations, regulatory officials from the Driver and Vehicle Licensing Authority (DVLA), Energy Commission, Environmental Protection Agency (EPA), and National Petroleum Authority (NPA), as well as LPG station attendants within the Kumasi Metropolis. The total population was estimated at approximately 15,000 individuals.

The representative sample size was determined using [22] formula for sample size estimation, ensuring adequate precision and confidence for population generalization. A total of 384 respondents were selected, proportionally distributed across the five study areas to capture diverse stakeholder perspectives.

3.3. Data Collection Instruments and Procedures

Primary data were collected using a structured questionnaire developed on a four-point Likert scale ranging from Strongly Agree to Strongly Disagree. The instrument was divided into thematic sections reflecting the study’s objectives, namely prevalence of LPG use, safety risks, environmental implications, regulatory challenges, and public awareness. Demographic information was also collected to enable subgroup analysis.

Data collection was conducted in person across the five identified areas. Respondents were approached directly and assisted in completing the questionnaires by the researcher and trained field assistants. Participation was voluntary, and informed consent was obtained from all respondents prior to data collection. To ensure data reliability and ethical integrity, respondents were assured of anonymity and confidentiality throughout the process.

The final sample comprised 384 valid responses from four stakeholder groups: vehicle operators, auto-technicians, regulatory officials, and LPG station attendants. Data collection spanned four weeks, ensuring comprehensive geographic and demographic coverage. The structured questionnaire design facilitated quantitative analysis using R Studio, version 2025.09.1, enabling computation of descriptive statistics and inferential tests aligned with the study’s analytical framework.

The research followed a sequential analytical workflow with mixed methods as shown in Figure 4. The process started with the development and management of a structured questionnaire to collect the basic data on LPG installation procedures, safety protocols and incident history of 384 participants in the Kumasi Incident.

The collected data were subsequently channelled into a two-pronged analytical framework to address the research objectives comprehensively:

• Multiple Correspondence Analysis (MCA) which is a multivariate technique was employed to reduce the dimensionality of the categorical survey data. The objective was to visualize the complex associations between variables (e.g., type of installer, safety equipment, incident type) and identify underlying patterns. The output of this analysis was a factor map that plotted categories and individuals, allowing for the emergent clustering of distinct user profiles based on their shared characteristics and experiences.
• The Random Forest model was trained using repeated k-fold cross-validation to ensure robust performance estimation and to minimize overfitting. The dataset of 384 observations was internally partitioned during the resampling process, with approximately 70% of cases used for training within each fold and 30% used for validation. Model performance was assessed using cross-validated confusion matrices, accuracy, sensitivity, specificity, and area under the Receiver Operating Characteristic (ROC) curve. No external holdout set was used; instead, repeated cross-validation provided stable and unbiased estimates of predictive performance.
• The results of both analytical streams were then synthesized and interpreted in parallel. Qualitative information from MCA user profiles provided a background for the quantitative predictions and ranking of variables in the Random Forest model. This integrated interpretation has provided a basis for robust conclusions and policy recommendations which can be implemented to mitigate the risks associated with non-standard LPG installations.

4. Results

4.1. Descriptive Results

Table 1. Demographic characteristics of respondents.

Variable	Category	Frequency	Percentage
Gender	Female	144	37.5
	Male	240	62.5
Age Range	18–25	85	22.1
	26–35	124	32.3
	36–45	92	24.0
	46+	83	21.6
Occupation	Business owner	71	18.5
	Auto-technician	41	10.7
	LPG attendant	33	8.6
	Regulatory official	35	9.1
	Taxi driver	124	32.3
	Trotro operator	80	20.8
Experience	<1 year	77	20.1
	1–3 years	98	25.5
	4–6 years	111	28.9
	>6 years	98	25.5

presents the demographic characteristics of the respondents. The results indicate that a majority (62.5%) of the participants were male, while females constituted 37.5%. Regarding age distribution, the 26–35 years category was the most represented (32.3%), followed by 36–45 years (24.0%) and 18–25 years (22.1%).

In terms of occupation, taxi drivers accounted for the largest proportion (32.3%), while LPG attendants represented the smallest (8.6%). The remaining respondents included trotro operators (20.8%), business owners (18.5%), regulatory officials (9.1%), and auto-technicians (10.7%). With respect to work experience, most participants (28.9%) had 4–6 years of experience, while 20.1% had less than one year of experience.

Table 2. Differences in Safety Practices and Incident Experience across Respondent Groups.

Risk Type	Business Owner	Auto-Technician	LPG Attendant	Regulatory Official	Taxi Driver	Trotro Operator	<1 Year	1–3 Years	4–6 Years	>6 Years	Female	Male
Fire Explosion	0.2676	0.1707	0.2121	0.1714	0.2097	0.1625	0.2597	0.1837	0.2162	0.1633	0.2222	0.1917
Gas Leak	0.3239	0.3415	0.3333	0.3429	0.2258	0.2500	0.2338	0.2347	0.3063	0.3367	0.2847	0.2792
None	0.2817	0.1951	0.3939	0.3714	0.3871	0.4625	0.3247	0.4184	0.3153	0.3878	0.3403	0.3750
Tank Failure	0.1268	0.2927	0.0606	0.1143	0.1774	0.1250	0.1818	0.1633	0.1622	0.1122	0.1528	0.1542

shows that there were evident and systematic differences in the experience of incident among respondent groups, which means that there was a significant difference according to occupation, level of experience, and gender. Auto-technicians had the largest percentage of tank failures at 29.27% as opposed to very low percentages of LPG attendants (6.06%) and regulatory officials (11.43%), indicating more technical exposure to cylinder parts. Gas leakage was the most frequently reported incident across nearly all groups, particularly among regulatory officials (34.29%) and auto-technicians (34.15%), while taxi drivers exhibited comparatively lower gas leak proportions (22.58%). Respondents with more than six years of experience reported the lowest proportions of fire or explosion incidents (16.33%) and tank failures (11.22%), alongside higher proportions of incident-free outcomes (38.78%), indicating a protective effect of experience. The differences between genders were small, with males reporting slightly higher proportions of no-incident outcomes (37.50%), while females reported marginally higher fire or explosion incidents (22.22%).

4.2. Multiple Correspondence Analysis (MCA)

The MCA results reveal that Dimension 1 (8.9%) captures a Safety and Regulation Compliance gradient, whereas Dimension 2 (6.0%) represents an Incident Experience or Risk gradient. Although the total explained variance (14.9%) appears modest, this is typical in MCA due to the inclusion of many categorical variables as demonstrated in Figure 5. Together, the first two dimensions provide a meaningful structure for interpreting the associations among safety practices, compliance levels, and incident outcomes.

In the top-left quadrant (positive Dim2, negative Dim1), an emerging tendency, labeled ‘At-Risk’, is observed among respondents associated with non-standard or self-installations, gas leaks, and fire/explosion incidents. These clusters represent tendencies in the data rather than strictly separated groups, acknowledging potential overlap across profiles. The bottom-left quadrant (negative Dim2, negative Dim1) represents a tendency towards professional installations and adherence to safety practices profile characterized by professional installations, certified technicians, and adherence to safety protocols, correlating with a lower likelihood of incidents.

The bottom-right quadrant (negative Dim2, positive Dim1) reflects an emerging pattern of equipment ownership and regulatory awareness profile, linking individuals who possess shut-off valves and fire extinguishers with support for stricter regulations.

The top-right quadrant (positive Dim2, positive Dim1) defines a tendency to formally report incidents and be aware of government policies profile—respondents aware of government policies and more likely to formally report incidents. This cluster’s proximity to “tank failure” variables suggests that some safety-oriented respondents may have adopted preventive measures following prior accidents.

To strengthen the interpretation of the MCA results, squared cosine values (cos²) and category contributions were examined to assess the quality of representation and the influence of variables on each dimension. From

Table 3. Categories with High cos² and Contributions in the MCA.

Category	cos2 (Dim 1)	cos2 (Dim 2)	Contribution (Dim 1)	Contribution (Dim 2)
Fire Extinguisher = Yes	0.0010	0.2318	0.06	13.04
Fire Extinguisher = No	0.0010	0.2318	0.02	5.04
Govt Policies = Aware	0.3319	0.1648	17.61	9.14
Govt Policies = Not Aware	0.3319	0.1648	7.16	3.72
Stricter Enforcement = Yes	0.3148	0.1076	16.95	6.05
Stricter Enforcement = No	0.3148	0.1076	6.55	2.34

, the results have shown that the government policy awareness and support of stricter implementation is highly represented in Dimension 1 and the cos² values are above 0.31 with high contributions as high as 17.61. This validates Dimension 1 as an axis of compliance and regulatory awareness, which is mostly motivated by the interaction of the respondents with formal policy framework and enforcement systems. The equal representation of the two categories of the negative and positive responses also indicates the fact that this dimension reflects a distinct distinction between a regulatory compliance and non- compliance orientations. On the other hand, the ownership of fire extinguishers has low values of cos² on Dimension 1 and high values of cos 2 on Dimension 2 at 0.232 with the highest contribution being given in that dimension at 13.04. This shows that the practical equipment of safety and exposure to accidents is more correlated with Dimension 2 than with regulatory knowledge. The overwhelming prevalence of safety equipment groups on Dimension 2 is what validates its perceived as a risk and response gradient, disclosing the behaviorally and experience-based differences in terms of the LPG safety practices.

The coordinate matrix (

Table 4. Coordinates of categories on the first five Multiple Correspondence Analysis (MCA) dimensions.

Row	Dim 1	Dim 2	Dim 3	Dim 4	Dim 5
1	0.027	0.592	−0.169	−0.077	−0.278
2	0.437	−0.133	−0.057	−0.391	−0.252
3	0.364	0.067	−0.296	0.185	−0.185
4	−0.345	−0.213	−0.223	0.216	−0.219
5	−0.020	−0.007	−0.538	−0.042	−0.175
6	−0.155	−0.114	−0.154	−0.213	−0.137

) confirms these patterns. High positive Dim2 scores (e.g., Row 1 = 0.592) align with incident-related experiences, while high positive Dim1 scores (e.g., Rows 2 and 3) correspond to safety compliance and regulatory alignment. Conversely, negative scores on both dimensions (Rows 4 and 6) denote proactive safety behaviour with minimal exposure to incidents. While Dimensions 1 and 2 account for the largest proportion of inertia and form the primary basis for interpretation, additional dimensions (Dimensions 3–5) were retained to capture residual but meaningful variability in safety practices and incident experiences. These higher dimensions represent secondary association structures and are reported for completeness rather than primary interpretation.

4.3. Random Forest Classification Results

4.3.1. Confusion Matrix and Accuracy

As shown in

Table 5. Confusion Matrix of Random Forest Classification.

Prediction	Fire Explosion	Gas Leak	None	Tank Failure
Fire Explosion	22	1	0	0
Gas Leak	0	30	0	0
None	0	0	41	1
Tank Failure	0	1	0	19

, the Random Forest model demonstrates exceptional classification accuracy across all four risk types. Most predictions align perfectly with actual outcomes, particularly for Gas Leak and None categories. The confusion matrix presented reflects aggregated predictions from cross-validated test folds rather than a single small holdout sample.

The Random Forest classifier achieved a predictive accuracy of 96.5% based on cross-validation, with high sensitivity and specificity at the chosen hyperparameter settings, with a 95% confidence interval between 91.2% and 99.0%. To reduce the risk of overfitting, the error convergence was monitored as the number of trees increased and performed repeated cross-validation. While the model shows strong predictive capability, results are presented as indicative of associations and patterns rather than definitive predictions (

Table 6. Overall Classification Statistics.

Metric	Value
Accuracy	0.9646
95% CI	(0.9118, 0.9903)
No Information Rate	0.3628
p-Value [Acc > NIR]	<2.2 × 10−16

).

4.3.2. Class-by-Class Performance

Table 7. Performance Statistics by Class.

Metric	Fire Explosion	Gas Leak	None	Tank Failure
Sensitivity	0.9565	0.9375	1.0000	0.9412
Specificity	0.9889	1.0000	0.9861	0.9792
Positive Predictive Value	0.9565	1.0000	0.9762	0.8889
Negative Predictive Value	0.9889	0.9759	1.0000	0.9895
Balanced Accuracy	0.9727	0.9688	0.9931	0.9602

presents class-level statistics. The model achieves excellent balance between sensitivity and specificity across all categories, confirming reliable discrimination among incident types.

The Random Forest error plot (Figure 6) illustrates how classification error stabilizes after approximately 150 trees, confirming model convergence and robustness. The figure is used to demonstrate the out-of-bag (OOB) error curves of the random-for est model when the ensemble size is varied by increasing the number of trees. The colored lines are used to depict the errors of prediction by classes or the general rate of error. The black solid line is a representation of the global OOB error of the model and it is closing in on about 0.68 at about 150 trees which is to say that further increase in the number of trees can only make it decrease by small margins and the model is approaching a point of diminishing returns. The dashed blue line represents the error rate of the class that has the highest misclassification probability. This line is also continuously on the high side of about 0.90 and that indicates that this particular class is the one that is significantly challenging to predict accurately using the current framework of the model. On the other hand, the dashed red line reflects a second mistake of the classes; it levels off a range of 0.78 to 0.80 thus indicating a moderate level of difficulty in prediction. There is a third class represented by the green and dotted line. Its error value is relatively low with a plateau of about 0.74 and the deviations are small around this plateau, which implies that this classification has a relatively higher performance than the red and light-blue classes. However, the dark blue, dotted line has a much smaller error rate: it falls drastically at the first 50 trees and levels out at 0.45. This trend suggests that this class is more easily learnt in the model compared to others.

4.3.3. Hyperparameter Tuning and Model Optimization

The cross-validated hyperparameter tuning results (

Table 8. Cross-Validated Hyperparameter Tuning Results.

Mtry	LogLoss	AUC	prAUC	Accuracy	Kappa	Mean F1
2	0.7118	0.999	0.933	0.966	0.953	0.960
3	0.5488	1.000	0.940	1.000	1.000	1.000
4	0.4410	1.000	0.940	1.000	1.000	1.000
5	0.3532	1.000	0.940	1.000	1.000	1.000
6	0.2820	1.000	0.942	1.000	1.000	1.000

) further validate the model’s robustness. As the number of predictors per split (mtry) increases from 2 to 6, the logLoss decreases steadily from 0.7118 to 0.2820, indicating better model fit. For mtry ≥ 3, both AUC and Accuracy reach 1.0, with Kappa and Mean F1 also achieving 1.0—demonstrating perfect classification and complete agreement beyond chance.

From mtry = 3 onward, mean sensitivity, specificity, and predictive values all reach 1.0, confirming that the model consistently identifies true cases while minimizing false positives and negatives.

As shown in Figure 7, all the four risk types, which are Fire Explosion, Gas Leak, None and Tank Failure, have perfect discriminatory ability using multi-class Receiver Operating Characteristic (ROC) curves. Every one of these class-specific ROC curves is entirely contained within the upper-left boundary of the plot, and has a total Area Under the Curve (AUC) equal to 1.00, so that the sensitivity and specificity are 100 per cent at all possible classification thresholds. The non-horizontal line marks the reference line of random classification, which is obviously dominated by the ROC curves of the model.

4.3.4. Variable Importance and Risk Factors

Table 9. Variable Importance and Key Factors Influencing LPG System Risks.

Variable	Fire Explosion	Gas Leak	None	Tank Failure
Type of Vehicle	−0.14	0.05	1.38	2.48
Installation Type	3.96	−3.21	−0.33	−3.67
Fuel Type	1.46	−0.35	1.74	−0.54
Safety Protocols	3.88	0.23	−2.22	3.65
Installer competence	2.09	−5.14	2.70	1.31
Service Frequency	−2.12	−4.51	6.69	4.36
Fire Extinguisher	−1.46	0.13	−0.76	−5.53
Government Policies	−1.24	3.86	1.52	0.92
Experience	−0.39	−3.01	0.54	−2.47
Occupation	−2.29	−4.19	5.33	5.37

highlights the most influential predictors of LPG-related incidents. Installation type, adherence to safety protocols, installer expertise, service frequency, and occupation exhibit strong associations with risk outcomes.

Poor installation practices significantly elevate fire and explosion risks, while regular servicing strongly enhances the likelihood of incident-free outcomes. Adherence to safety protocols may reflect post-incident adaptations, where safety measures were implemented after prior incidents. In contrast, variables such as installer competence, installation type, and service frequency represent pre-incident risk factors that influence the likelihood of incidents occurring. Separating these categories helps to better understand drivers of risk and guide targeted interventions. Occupation and installer competence emerge as dual-effect variables: safer handling among professionals reduces explosion and leak probabilities but may still expose vehicles to tank failures due to high operational frequency.

5. Discussion

The study set out to examine the prevalence, safety implications, and risk determinants of non-standard LPG vehicle conversions in the Kumasi Metropolis, Ghana. Findings from the descriptive and predictive analyses underscore both the increasing adoption of LPG as an automotive fuel and the significant safety and regulatory challenges associated with its unstandardized use. The discussion integrates these empirical findings with existing literature to provide a contextualized understanding of LPG conversion practices and associated risks in developing urban transport systems.

5.1. Prevalence and Drivers of LPG Conversion

Results from the field survey revealed that a substantial proportion of commercial and private drivers in Kumasi have converted their vehicles to run on LPG, primarily as a cost-saving response to rising petrol and diesel prices. This finding corroborates observations by [23] who reported similar economic motivations driving LPG adoption among transport operators in Nigeria and Ghana, respectively. The widespread informal conversions observed—especially within Suame Magazine and Asafo—reflect a growing informal energy transition that is largely market-driven rather than policy-led.

Unlike in industrialized settings where LPG conversion is regulated and technically standardized, the Ghanaian context is characterized by fragmented oversight, limited enforcement, and informal technical practices. Consequently, cost minimization often takes precedence over adherence to safety standards, aligning with [5,16] who found that economic incentives often overshadow regulatory compliance in sub-Saharan African vehicle energy transitions.

5.2. Safety Risks and Technical Compliance

The MCA results provide more information into the behavioural and regulatory dimensions underlying LPG safety outcomes. Dimension 1 clearly represents a regulatory compliance and policy awareness gradient, strongly driven by awareness of government policies and support for stricter enforcement. This dimension separates respondents who actively engage with formal regulatory frameworks from those operating largely outside institutional oversight.

Dimension 2 captures an incident exposure and risk-response gradient, characterized by associations with gas leaks, fire/explosion incidents, and ownership of safety equipment such as fire extinguishers. The prominence of safety equipment on this dimension suggests that safety measures are often adopted reactively rather than proactively, following prior exposure to incidents.

The quadrant-based interpretation further reveals distinct behavioural tendencies. The “at-risk” profile is associated with non-standard or self-installations and higher incident occurrence, highlighting the dangers of informal installation practices. Conversely, the professional installation and compliance-oriented profile align with lower incident exposure, emphasizing the protective role of certified installers and adherence to safety protocols. Interestingly, the cluster associated with policy awareness and formal reporting is positioned close to tank failure outcomes, suggesting that some respondents become more compliant and safety-conscious after experiencing serious incidents, rather than beforehand.

The Random Forest model developed in this study identified critical predictors of safety risk, achieving excellent classification performance (AUC = 1.0, accuracy = 1.0 at mtry ≥ 3). This robust predictive capability suggests that risk factors are well-structured and discernible based on key attributes such as conversion site, technician qualification, inspection frequency, and tank age. Consistent with [17,19] their results indicate that unprofessional installations, absence of regulatory inspection, and use of expired or uncertified cylinders are strong determinants of high safety risk.

These findings reinforce the position that informal conversion practices substantially elevate accident likelihoods, fire hazards, and emissions irregularities. Evidence from similar studies in [5] also highlight that informal fuel conversions often bypass essential leak tests, pressure ratings, and component certifications, resulting in elevated risk exposure for both users and the public.

Furthermore, the high sensitivity and specificity values obtained demonstrate that the identified risk features can reliably distinguish between safe and unsafe conversion practices.

5.3. Regulatory Gaps and Institutional Challenges

Evidence from key informant interviews and field observations revealed that regulatory bodies such as the DVLA, EPA, Energy Commission, and NPA have overlapping but weakly coordinated mandates regarding vehicular LPG regulation. These perception of the institutional inefficiencies mirror findings by [5,9] who emphasized fragmented institutional structures as major barriers to energy safety governance in Ghana. While the NPA oversees LPG marketing and storage, the DVLA’s vehicle inspection mandate does not extend to post-conversion verifications, creating a regulatory void that informal installers exploit.

This lack of synergy contributes to the persistence of unapproved installations, absence of certification procedures, and low enforcement of existing standards.

In contrast, jurisdictions of some European countries have established unified licensing and inspection frameworks that integrate vehicle and fuel system regulation [24], demonstrating that coordinated institutional mechanisms can effectively mitigate conversion-related safety risks.

5.4. Environmental Implications

From an environmental perspective, the shift toward LPG use represents a partial transition to cleaner energy within the transport sector. LPG combustion emits significantly lower particulate matter and CO₂ compared to gasoline and diesel. However, the environmental benefit diminishes when conversions are technically substandard or when systems leak, resulting in unburnt hydrocarbon emissions. This aligns with [4,25], who found that improper retrofitting can negate the expected emission benefits of LPG adoption. Thus, while the proliferation of LPG use in Kumasi contributes to fossil fuel substitution, the environmental gains remain conditional on regulatory enforcement and conversion quality.

5.5. Policy and Practical Implications

The empirical and predictive evidence suggests that informal LPG conversion practices present a dual challenge: they facilitate cost savings and partial decarbonization but simultaneously increase safety vulnerabilities. Addressing this paradox requires a multi-level policy response. First, standardization and certification of conversion centres should be prioritized, accompanied by mandatory technician training and licensing. Second, a centralized inspection and reporting system linking the DVLA, Energy Commission, and NPA could close existing regulatory gaps. Finally, public awareness campaigns on safe LPG handling and proper system maintenance should be institutionalized through transport unions and local government assemblies.

5.6. Limitations and Future Research

This study has limitations. Its cross-sectional design infers association but not causation. The reliance on self-reported data may be subject to bias. Future longitudinal studies could track incident rates following policy interventions. Furthermore, technical inspections of a sub-sample of vehicles could provide objective data to correlate with the survey responses.

5.7. Theoretical Contribution

Integrating predictive analytics (Random Forest classification) with field-based survey data offers a novel methodological contribution to transport safety research in low- and middle-income countries. The model indicates that machine learning techniques can identify important associations and patterns among risk factors in informal LPG conversion practices, providing a data-driven basis for understanding potential safety risks, rather than asserting perfect predictive capability.

This integrated approach responds to calls by [26] for the application of artificial intelligence techniques to enhance safety management in decentralized energy systems. It further extends the theoretical discourse on informal energy transitions by empirically linking behavioural, technical, and institutional determinants within a predictive framework.

6. Conclusions

This study examined the prevalence, safety risks, and regulatory dynamics of Liquefied Petroleum Gas (LPG) vehicle conversions in the Kumasi Metropolis of Ghana using both empirical and predictive analytical frameworks. The findings revealed that LPG has emerged as a cost-effective fuel alternative among commercial and private drivers, primarily driven by economic pressures associated with rising petroleum prices. However, the expansion of this informal fuel transition is accompanied by significant safety and regulatory challenges.

Descriptive and predictive analyses demonstrated that non-standard conversions, inadequate inspections, and the use of uncertified components constitute major safety risk determinants. The Random Forest model achieved perfect or near-perfect classification performance (AUC = 1.0, accuracy = 1.0 for mtry ≥ 3), confirming that risk factors are well-structured and can be reliably predicted based on technician qualification, conversion site, and maintenance practices. These results highlight the utility of data-driven approaches in assessing and managing safety risks associated with informal energy practices.

The study further established that institutional fragmentation among key agencies—including the DVLA, Energy Commission, Environmental Protection Agency, and National Petroleum Authority—creates regulatory gaps that undermine safety enforcement. Despite LPG’s environmental advantages over petrol and diesel, the expected emission reduction benefits are compromised by unregulated retrofitting, leakage, and poor maintenance. Thus, the study concludes that while LPG adoption supports Ghana’s broader energy diversification agenda, its sustainability depends on strengthening institutional coordination, enforcing safety standards, and promoting technical compliance.

6.1. Policy Recommendations

1.

Standardization and Licensing of Conversion Centres

Establish a national registry of certified LPG conversion workshops under the joint supervision of the Energy Commission and DVLA. Only licensed technicians with verifiable training should be authorized to perform LPG installations and maintenance.

2.

Integrated Inspection and Data Management System

Develop a centralized digital platform linking the DVLA, NPA, and Energy Commission to record all LPG vehicle conversions, inspection results, and certification renewals. Such integration would enhance traceability and facilitate risk-based regulatory interventions.

3.

Capacity Building and Technician Certification

Implement mandatory training and continuous professional development programs for auto-technicians in collaboration with technical institutes and transport unions. Certification should be competency-based and periodically renewed to ensure adherence to safety standards.

4.

Public Awareness and Driver Education Campaigns

Launch sustained educational initiatives on LPG safety, proper maintenance, and early fault detection. Collaboration with transport unions and local assemblies will help disseminate safety information effectively at the grassroots level.

5.

Periodic Safety Audits and Enforcement Mechanisms

Conduct annual joint inspections of LPG-converted vehicles and fuelling stations to identify non-compliance. Penalties and license revocations should be enforced to deter unsafe practices while incentivizing compliance through certification benefits.

6.

Promotion of Clean Fuel Transitions within Formal Frameworks

Position LPG adoption within Ghana’s clean energy policy framework by integrating it into the National Energy Transition Plan. This approach will align fuel diversification with decarbonization objectives, ensuring that LPG serves as a transitional rather than informal energy solution.

6.2. Future Research Directions

Future studies should explore longitudinal data on LPG vehicle performance and accident records to quantify the actual safety and environmental impacts over time. Expanding predictive modelling to include other urban centres such as Accra and Takoradi would facilitate cross-regional comparison and enhance generalizability. Integrating behavioural factors, such as risk perception and compliance motivation—into predictive frameworks could also provide deeper insight into driver decision-making and improve policy targeting.