Project – Predictive Maintenance of Urban Road Pavements Using Machine Learning and Pavement Distress Indicators. A Study of Selected urban roads in Kaduna metropolis

CHAPTER ONE

INTRODUCTION

1.1 Background to the Study

Road infrastructure is a fundamental component of urban development because it facilitates the movement of people, goods, and services and supports commercial activities, employment, and access to essential facilities. The condition of road pavements influences travel comfort, road safety, vehicle operating costs, and the efficiency of urban transportation systems. As road pavements are exposed to repeated traffic loading, environmental changes, water infiltration, and material deterioration, their structural and surface conditions gradually change. If these changes are not detected and addressed at the appropriate time, minor pavement defects may develop into more severe forms of deterioration, increasing the cost and complexity of subsequent repairs. The American Association of State Highway and Transportation Officials (AASHTO, 2012) emphasises the importance of systematic pavement management in evaluating pavement condition, forecasting future performance, and selecting appropriate maintenance and rehabilitation strategies. Similarly, the Federal Highway Administration (FHWA, 2014) identifies pavement condition assessment and performance monitoring as important components of effective pavement asset management. These principles establish the need for road authorities to move beyond reactive repairs towards systematic approaches that identify deterioration early and support timely maintenance decisions.

Traditional road pavement maintenance often depends on periodic visual inspections, reported road defects, and the professional judgement of engineers and maintenance personnel. Although these methods remain important, they may be limited by the frequency of inspection, the availability of trained personnel, inconsistent defect recording, and the difficulty of monitoring extensive road networks continuously. Pavement distress can take several forms, including longitudinal and transverse cracking, fatigue or alligator cracking, potholes, rutting, ravelling, bleeding, and surface deformation. Each distress type provides information about the possible condition and performance of a pavement, although its significance depends on its severity, extent, location, and the pavement’s structural and environmental characteristics. ASTM International’s standard practice for pavement condition surveys provides a structured procedure for identifying and recording pavement distress and calculating a Pavement Condition Index (PCI) (ASTM International, 2024). A consistent assessment system can support comparisons between road sections and help maintenance agencies identify locations requiring further investigation. However, a condition survey that describes the current state of a pavement does not necessarily predict how rapidly it will deteriorate or when maintenance should be undertaken. This distinction makes predictive analysis an important extension of conventional pavement condition assessment.

Predictive maintenance is an approach in which maintenance activities are planned using evidence about the current condition of an asset and estimates of its likely future performance. In road engineering, predictive maintenance involves analysing pavement distress indicators and other relevant variables to estimate future deterioration, identify roads at risk of reaching unacceptable condition levels, and determine suitable intervention priorities. This approach differs from purely reactive maintenance, in which repairs are initiated after defects become apparent or complaints are received, and from time-based maintenance, in which activities are scheduled primarily according to fixed intervals. A pavement management system provides an organised framework for collecting road condition information, assessing deterioration, forecasting future performance, and comparing maintenance alternatives. The FHWA (2014) explains that pavement management relies on systematic condition information and performance prediction to support decisions concerning maintenance and rehabilitation. Machine learning can strengthen this process by identifying patterns within historical inspection records and relating observed distress to factors such as pavement age, traffic loading, drainage conditions, and previous maintenance. Consequently, predictive maintenance offers a potential means of improving the timing and prioritisation of pavement interventions, provided that the models are developed and validated using suitable data.

Machine learning is a branch of artificial intelligence that enables computational systems to identify patterns in data and generate predictions or classifications without relying exclusively on manually specified relationships. Its application to pavement management has attracted attention because pavement deterioration is influenced by multiple interacting factors and may not follow a simple linear pattern. Sholevar et al. (2022), in their review of machine-learning techniques for pavement condition evaluation, examined methods for image classification, distress detection, image segmentation, and pavement condition assessment. Their review highlights the growing role of data-driven methods in processing pavement information and the continuing challenges associated with identifying complex distress patterns and estimating their severity and density. Machine-learning methods such as decision trees, random forests, support vector machines, gradient boosting, and neural networks can be trained to classify pavement condition or predict numerical indicators from available observations. The suitability of a particular algorithm depends on the quality and quantity of data, the prediction objective, the distribution of pavement conditions, and the need for interpretable results. For example, a classification model may estimate whether a road section is in good, fair, or poor condition, while a regression model may predict a numerical pavement condition score or the future severity of a specific distress. These capabilities make machine learning a potentially valuable tool for supporting proactive road maintenance.

Pavement distress indicators provide the measurable inputs required to develop and evaluate predictive maintenance models. Such indicators may include the type of distress, its severity, affected area or length, pothole count, rut depth, pavement roughness, surface condition, and overall pavement condition index. Depending on data availability, additional explanatory variables may include road age, traffic volume, the proportion of heavy vehicles, drainage characteristics, rainfall, pavement structure, and previous maintenance history. Sholevar et al. (2022) explain that pavement condition evaluation increasingly involves automated data collection and computational analysis, while ASTM International (2024) provides a standardised basis for identifying and quantifying pavement distress during condition surveys. Recent empirical research has demonstrated the feasibility of using machine-learning models to predict pavement distress indicators from road-surface inspection data. For example, a 2025 study published in Case Studies in Construction Materials compared several regression algorithms for predicting cracking, rutting, aggregate stripping, and other pavement characteristics and reported strong performance from ensemble methods, including random forest and gradient boosting, for selected indicators (A Case Study on Predicting Asphalt Pavement Distress Using Advanced Machine Learning Techniques and Road Surface Inspection Data, 2025). These findings support the potential of machine learning for pavement prediction, but they also reinforce the importance of selecting suitable distress variables, evaluating prediction errors, and validating model performance before using predictions for maintenance decisions.

The adoption of predictive maintenance has particular importance in urban areas, where roads experience different traffic intensities, land-use pressures, drainage conditions, and maintenance histories. Roads serving commercial districts, markets, public institutions, residential neighbourhoods, and major transport corridors may deteriorate at different rates even when they were constructed using similar materials. Frequent heavy-vehicle movements can contribute to rutting and fatigue-related distress, while inadequate drainage can allow water to weaken pavement layers and accelerate deterioration. Variations in road construction quality, pavement age, maintenance history, and environmental exposure may also contribute to differences in condition. Machine-learning models can potentially capture these relationships when the relevant variables are available in a sufficiently representative dataset. However, models developed in one geographical area may not perform equally well in another because traffic patterns, pavement materials, climate, data collection practices, and maintenance standards may differ. Tamagusko et al. (2024), in their review of machine-learning applications in road pavement management, discuss the opportunities and implementation challenges associated with integrating data-driven methods into pavement management systems. This highlights the need to develop and evaluate predictive approaches using data that reflect the specific characteristics of the road network being studied.

Kaduna metropolis provides a relevant setting for investigating the application of machine learning to urban pavement maintenance. As a major urban centre in north-western Nigeria, Kaduna accommodates residential, commercial, institutional, and industrial activities that depend on an effective road network. The condition of its urban roads has implications for daily commuting, access to public services, commercial movement, and the cost of operating vehicles. The selection of urban roads for a predictive maintenance study creates an opportunity to examine whether recorded pavement distress indicators can be used to classify current pavement condition and estimate maintenance needs more systematically. The proposed study will focus on selected urban roads in Kaduna metropolis and investigate how machine-learning methods can be applied to pavement distress data to support condition prediction and maintenance prioritisation. The model may use variables such as cracking, potholes, rutting, surface deformation, pavement condition scores, and other relevant road characteristics, depending on the data collected. By evaluating the predictive performance of selected algorithms against observed pavement conditions, the study seeks to establish whether a data-driven approach can provide useful and locally relevant evidence for planning pavement maintenance. Its purpose is not to assume that machine learning will automatically reduce maintenance costs or prevent pavement failure, but to assess its practical value as a decision-support tool for urban road management.

1.2 Statement of the Problem

The maintenance of urban road pavements is a continuing challenge for road authorities because pavements deteriorate progressively under traffic loading, environmental exposure, drainage problems, and other operational conditions. When deterioration is not identified and addressed early, minor defects such as surface cracking and small potholes may develop into more serious damage, potentially requiring expensive rehabilitation or reconstruction. Traditional maintenance practices that depend mainly on periodic inspection and visible defects may not provide sufficient information about the likely future condition of every road section. Although systematic pavement condition surveys can help identify existing defects, they do not by themselves establish how quickly those defects will develop or which road sections are most likely to require intervention next. The Federal Highway Administration (FHWA, 2014) identifies pavement condition assessment and performance prediction as important components of pavement management. The problem, therefore, is how to use available pavement condition information to anticipate deterioration and support timely maintenance decisions instead of depending primarily on reactive repairs.

A second problem concerns the difficulty of predicting pavement deterioration from multiple distress indicators and influencing factors. Cracking, potholes, rutting, surface deformation, and roughness may occur in different combinations and at different levels of severity. Their development may also be influenced by pavement age, traffic loading, drainage, material characteristics, weather conditions, and previous repairs. As a result, a single distress indicator may not adequately represent the overall condition or future maintenance requirements of a road section. Sholevar et al. (2022) identify machine learning as an important area of pavement condition evaluation while noting continuing challenges in detecting complex distress patterns and determining their severity and density. Without a reliable method for combining relevant distress indicators, maintenance planners may find it difficult to compare road sections objectively, predict which sections are likely to deteriorate, and allocate limited maintenance resources according to risk and need.

A third problem is the difficulty of translating advances in machine learning into reliable, locally appropriate maintenance decisions. Although machine-learning models have demonstrated potential for pavement distress prediction, their performance depends on the relevance, completeness, and representativeness of the training data. Models trained on road networks with different pavement materials, traffic conditions, climatic exposure, and inspection procedures may not produce equally reliable predictions when applied elsewhere. Tamagusko et al. (2024) discuss both the opportunities and challenges associated with machine learning in road pavement management, including the need to address practical implementation issues. Furthermore, a model that performs well on one dataset may provide unreliable results for road sections or distress types that are poorly represented in its training data. The problem is therefore not merely to select a machine-learning algorithm, but to establish which model can predict pavement condition reliably from the distress indicators and road characteristics available for the selected urban roads in Kaduna metropolis.

The central problem addressed by this study is the need for a systematic, data-driven approach to predicting pavement condition and supporting maintenance prioritisation on selected urban roads in Kaduna metropolis. Without a locally evaluated predictive framework, maintenance decisions may continue to rely heavily on current visible defects, periodic assessments, or subjective judgements, which may not fully capture the future deterioration risk of each road section. This study will address the problem by analysing selected pavement distress indicators, developing and comparing appropriate machine-learning models, and evaluating their predictive performance using suitable measures. The resulting predictions may then be used to classify pavement condition or identify road sections that require closer inspection and possible maintenance intervention. The study will not assume that a machine-learning model can replace engineering inspections; rather, it will investigate whether machine learning can complement conventional pavement assessment and provide useful evidence for more consistent and proactive maintenance planning.

1.3 Aim and Objectives of the Study

The aim of this study is to develop and evaluate a machine-learning approach for predicting urban road pavement condition and supporting predictive maintenance decisions using pavement distress indicators on selected urban roads in Kaduna metropolis.

The specific objectives are to:

1. Identify and assess the major pavement distress indicators and road characteristics affecting the condition of selected urban roads in Kaduna metropolis.

2. Develop machine-learning models for predicting pavement condition using the identified distress indicators and relevant road characteristics.

3. Evaluate and compare the predictive performance of selected machine-learning algorithms using appropriate model evaluation metrics.

4. Develop a predictive maintenance prioritisation framework for identifying road sections that require monitoring, routine maintenance, or more substantial intervention based on predicted pavement condition.

1.4 Research Questions

The study will answer the following research questions:

1. What are the major pavement distress indicators and road characteristics affecting the condition of selected urban roads in Kaduna metropolis?

2. How can machine-learning models be developed to predict pavement condition using the identified distress indicators and relevant road characteristics?

3. Which of the selected machine-learning algorithms provides the most reliable predictions of pavement condition?

4. How can predicted pavement conditions be used to prioritise road sections for monitoring and maintenance interventions?

1.5 Research Hypothesis

The following null hypothesis will be tested at the 0.05 level of significance:

H₀: There is no statistically significant difference in the predictive performance of the selected machine-learning models for estimating pavement condition on selected urban roads in Kaduna metropolis.

The hypothesis will be tested by comparing the performance of the selected algorithms on the same held-out test dataset. The choice of statistical test will depend on the model comparison design and whether the prediction errors meet the assumptions of the selected test.

1.6 Significance of the Study

The findings of this study are expected to be useful to road maintenance authorities, civil engineers, transport planners, researchers, and other stakeholders responsible for urban road infrastructure.

Road maintenance authorities: The study may provide an evidence-based method for identifying road sections that are likely to experience poor pavement conditions. Predictions may help authorities rank maintenance needs and direct limited resources towards locations requiring timely attention.

Civil and pavement engineers: The study may support engineers in examining relationships between distress indicators and overall pavement condition. It may also help them compare machine-learning predictions with conventional pavement condition assessments when planning inspections and maintenance.

Kaduna State road management agencies: The findings may provide locally relevant evidence for improving road-condition monitoring and maintenance planning. A validated predictive model could complement existing inspection procedures and help organise road sections into maintenance priority categories.

Road users and the general public: Better-informed maintenance decisions may contribute to improved pavement conditions, more comfortable journeys, reduced vehicle damage, and safer road use. These benefits would depend on the extent to which model predictions are translated into appropriate maintenance actions.

Researchers and students: The study may contribute to research on machine learning, pavement distress assessment, infrastructure asset management, and predictive maintenance. It may also provide a basis for further investigations involving larger datasets, automated image-based distress detection, and longitudinal pavement deterioration forecasting.

Policy and infrastructure planners: The study may help demonstrate how data-driven methods can complement conventional engineering judgement in managing urban infrastructure. The findings may support discussions about digital road-condition monitoring, evidence-based maintenance prioritisation, and the gradual adoption of predictive pavement management systems.

1.7 Scope of the Study

The study will focus on selected urban roads within Kaduna metropolis, Kaduna State, Nigeria. The specific roads will be chosen based on accessibility, the availability of pavement condition information, the presence of observable distress indicators, and the feasibility of conducting field assessments.

The study’s subject scope covers pavement distress identification, pavement condition prediction, machine-learning model development, and predictive maintenance prioritisation. The distress indicators may include longitudinal and transverse cracking, alligator cracking, potholes, rutting, ravelling, surface deformation, and other relevant pavement defects. Where available, additional variables such as pavement age, traffic volume, drainage condition, road classification, and previous maintenance history may be included.

The methodological scope will involve collecting or compiling pavement condition data, preparing a dataset, selecting appropriate machine-learning algorithms, training and testing the models, and evaluating their predictive performance. Depending on the available data, the algorithms may include decision trees, random forests, support vector machines, and gradient boosting. Suitable metrics may include accuracy, precision, recall, and F1-score for classification tasks, or mean absolute error, root mean squared error, and the coefficient of determination for numerical prediction tasks.

The final stage will involve using the model’s predictions to propose a maintenance prioritisation framework. The model will serve as a decision-support tool rather than a replacement for engineering inspection, structural assessment, or professional judgement.

1.8 Limitations of the Study

The study may be affected by limitations in the availability, completeness, and consistency of pavement condition data. If historical inspection records are incomplete, the model may have insufficient information to learn how pavement distress develops over time. The researcher should document data gaps and avoid making deterioration forecasts that require longitudinal information when only a single inspection is available.

Another limitation may arise from variations in how pavement distress is identified and measured. Differences in inspection methods, severity ratings, road-section lengths, and recording procedures may affect the consistency of the dataset. The use of standardised inspection procedures and clearly defined distress categories can help reduce this problem.

The study may also be constrained by the number of road sections available for analysis and the representation of different pavement conditions in the dataset. If poor-condition roads or particular distress types are underrepresented, model predictions for those categories may be less reliable. Model validation should therefore consider class imbalance and the performance of the models across different pavement-condition categories.

Finally, the predictive performance observed during model testing may not fully reflect performance on roads that were not represented in the original dataset. The results should consequently be interpreted within the context of the selected urban roads in Kaduna metropolis. Wider application would require additional validation using data from other roads and periods.

1.9 Operational Definition of Terms

Predictive maintenance: A maintenance approach that uses information about an asset’s current condition and predicted future performance to identify when maintenance or intervention may be required.

Urban road pavement: The constructed road surface and supporting pavement layers within an urban road network, designed to carry traffic under specified loading and environmental conditions.

Machine learning: A branch of artificial intelligence in which computational algorithms learn patterns from data and use those patterns to generate predictions or classifications.

Pavement distress: Observable deterioration or damage affecting a road pavement, including cracking, potholes, rutting, ravelling, and surface deformation.

Pavement distress indicators: Measurable characteristics describing the type, extent, severity, or frequency of pavement defects and used to assess or predict pavement condition.

Pavement Condition Index (PCI): A numerical indicator of pavement surface condition calculated from recorded distress types, severities, and quantities using a defined assessment procedure.

Pavement condition prediction: The estimation of a pavement’s condition category or numerical condition measure using relevant distress data and a predictive model.

Predictive maintenance prioritisation: The process of ranking road sections according to their predicted condition, deterioration risk, and maintenance needs to support intervention planning.

Model accuracy: A measure of the proportion of correctly predicted cases in a classification task. Accuracy should be interpreted alongside other metrics when the pavement-condition categories are imbalanced.

Random forest: A machine-learning algorithm that combines predictions from multiple decision trees to produce a classification or regression result.

Urban road network: A connected system of roads serving movement within an urban area, including residential streets, collector roads, arterial roads, and other urban routes.

Kaduna metropolis: The urban study area within Kaduna State, Nigeria, from which the selected road sections for this research will be identified.

Project – Predictive Maintenance of Urban Road Pavements Using Machine Learning and Pavement Distress Indicators. A Study of Selected urban roads in Kaduna metropolis