Pothole Prediction Based on Machine Learning and Pavement Condition Indicators - Road Network Implementation Recommendations

Aug 5, 2026

Upload by Skyler

Potholes are hazardous road defects that contribute to vehicle, motorcycle, and bicycle crashes while creating a substantial financial load for highway authorities. Current approaches lack effective tools to predict both the number and spatial location of potholes across road networks. This study develops pothole prediction tools using Random Forest and K Nearest Neighbour, relying on nine SCANNER-derived pavement condition indicators and training on a Transport for London dataset, achieving high section-level detection rates and actionable guidance to improve accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Abstract

Introduction

Background and impacts of potholes

Existing survey and detection approaches

Proposed machine learning prediction approach

Required pavement condition indicators

Dataset and validation overview

Results and performance assessment

Section-level prediction outcomes

Recommendations and applications

Road authorities and pavement management planning

1 Pothole Prediction Based on Machine Learning and Pavement 2 Condition Indicators 3 4 Ahmed Abed, PhD, FIHE, FHEA, CEng 5 Department of Civil Engineering 6 Aston University, Birmingham, B4 7ET, UK 7 Corresponding author: [abeda@aston.ac.uk](abeda@aston.ac.uk) [8](8 Correspondence address:)[ Correspondence address:](8 Correspondence address:) 9 MB222D 10 Main Building 11 Aston University 12 Birmingham 13 UK 14 B4 7ET 15 ORCID ID: 0000-0002-6822-5519 16 17 Mujib Rahman, 18 Department of Civil Engineering 19 Aston University, Birmingham, B4 7ET, UK. 20 [rahmam19@aston.ac.uk](rahmam19@aston.ac.uk) 21 22 Nick Thom, MA, PhD, MICE, MCIHT, MPWI, CEng 23 Nottingham Transportation Engineering Centre 24 University of Nottingham, Nottingham, NG7 2RD, UK 25 [nicholas.thom@nottingham.ac.uk](nicholas.thom@nottingham.ac.uk) 26 27 David Hargreaves, 28 Faculty of Engineering 29 University of Nottingham, Nottingham, NG7 2RD, UK 30 [david.hargreaves@nottingham.ac.uk](david.hargreaves@nottingham.ac.uk) 31 32 Linglin Li, PhD, FIHE, 33 Nottingham Transportation Engineering Centre 34 University of Nottingham, Nottingham, NG7 2RD, UK 35 [linglin.li@nottingham.ac.uk](linglin.li@nottingham.ac.uk) 36 37 38 Gordon Airey, PhD, MCIHT 39 Nottingham Transportation Engineering Centre 40 University of Nottingham, Nottingham, NG7 2RD, UK 41 [gordon.airey@nottingham.ac.uk](gordon.airey@nottingham.ac.uk) 42 43 44 Number of words: 5414 45 46 47 1 Abstract 2 3 Potholes are dangerous defects on road surfaces, contributing to numerous 4 crashes involving vehicles, motorcycles, and bicycles. They also impose a significant 5 economic burden on highway authorities. Currently, no effective tool predicts the 6 number and location of potholes in a road network. In this study, an attempt to 7 address this gap has been proposed by developing novel pothole prediction tools 8 built using two machine learning methods, Random Forest, and K Nearest 9 Neighbour. The final prediction model requires nine pavement condition indicators, 10 quantifiable through Surface Condition Assessment for the National Network of 11 Roads (SCANNER) surveys, commonly conducted in the UK. This unique approach 12 allows for direct implementation by highway authorities. The model has been trained 13 on a large dataset of pavement condition and pothole data covering the ‘Transport 14 for London’ network. Validation results suggest that the model successfully predicted 15 55.5% of sections with potholes and 99.6% of sections without potholes. Although 16 the model demonstrates limited accuracy in predicting potholes, recommendations 17 are provided to enhance its performance. This work holds the potential to 18 significantly aid strategic financial planning for pavement management in the UK 19 and beyond. 20 21 Keywords: Pavements and Roads, Modelling, Potholes; Pavement Condition; 22 SCANNER, Machine Learning, UN SDG 9. 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 1 Introduction 2 Potholes are localised defects that form in the surface of roads usually having a 3 bowl shape with varying widths, lengths and depths (Miller and Bellinger, 2003) . In 4 recent years, this distress has attracted considerable attention by the public in the 5 UK due to the ever-rising number of potholes forming in UK roads, and the critical 6 negative economic, social, and environmental impacts this distress causes. The latest 7 Annual Local Authority Road Maintenance (ALARM) survey report shows that 8 1,985,480 potholes were filled in 2023 by the local authorities in England and Wales 9 at an average cost of £72.27 per pothole (AiA, 2024) . This represents a 40% increase 10 in the number of filled potholes and a 14.4% increase in the average pothole filling 11 cost compared to 2022 . Bearing i

FAQs

What problem does the study address in pothole management?

The study addresses the lack of effective tools that can predict the number and location of potholes across a road network. It aims to support more reliable planning for pavement management.

Which machine learning methods are used for pothole prediction?

The approach uses two machine learning methods: Random Forest and K Nearest Neighbour. Both are combined with nine pavement condition indicators derived from SCANNER surveys.

How were prediction performance results evaluated and what do they indicate?

Validation indicates the model predicted 55.5% of sections with potholes and 99.6% of sections without potholes. The work also notes limited accuracy for pothole prediction and provides recommendations to improve performance.

Aug 5, 2026

Upload by Skyler