🛣️ Machine Learning for Pavement Deterioration & Roughness (IRI) Prediction

---

📌 Executive Summary

Predicting highway pavement deterioration is critical for transportation asset management, road safety, and optimizing municipal maintenance budgets.

This repository presents a machine learning framework to predict pavement surface roughness—measured as the International Roughness Index (IRI)—using field-measured Cumulative Traffic Volume (AADT) and pavement Effective Structural Capacity (Total SN).

The models are trained, cross-validated, and benchmarked on real-world field survey data collected across the National Highway N-5 corridor (Sangjani Toll Plaza to Attock Bridge, Pakistan). The repository includes:

• Cleaned Field Datasets: 75 one-kilometer highway segments with classified traffic counts and structural deflection backcalculations.
• Comparative ML Benchmarks: Linear Regression, Decision Trees, Random Forests, Gradient Boosting, and a domain-specific piecewise decision model.
• Interactive Prediction Tools: Python CLI predictor (src/predict.py), a Jupyter Notebook, and an Excel macro deployment (.xlsm).
• Pavement Condition Classification: Automated mapping of predicted roughness to Federal Highway Administration (FHWA) and World Bank maintenance categories.

---

📊 Dataset & Field Engineering Methodology

The dataset encompasses 75 highway segments ($1\text{ km}$ each) surveyed along the N-5 Grand Trunk Road.

1. Highway Sections Surveyed

1. Tarnol to Taxila ($AADT = 4,133\text{ veh/day}$)
2. Taxila to Hassanabdal ($AADT = 6,796\text{ veh/day}$)
3. Hassanabdal to Burhan ($AADT = 2,852\text{ veh/day}$)
4. Burhan to Kamrah ($AADT = 934.5\text{ veh/day}$)
5. Kamrah to Attock Bridge ($AADT = 1,902\text{ veh/day}$)

2. Feature Engineering & Parameters

• Independent Feature 1: Effective Structural Number ($SN_{eff}$ / Total $SN$)
  Derived from Heavy Weight Deflectometer (HWD) surface deflection basins ($d_0$ in mils). Represents the structural stiffness and layer thickness capacity of the asphalt, base, and subgrade layers.
• Independent Feature 2: Annual Average Daily Traffic ($AADT$)
  Classified vehicle counts including heavy commercial vehicles, buses, multi-axle trucks, and passenger cars.
• Target Variable: International Roughness Index ($IRI$)
  Recorded via road surface profilometer. Quantified in units of $\text{in/mi}$ (and converted to $\text{m/km}$).

---

🔬 Machine Learning Benchmarking

Five modeling approaches were benchmarked on the 75 highway segments:

Model Architecture	MAE (in/mi)	RMSE (in/mi)	$R^2$ Score	Key Characteristics
Gradient Boosting Regressor	15.16	22.52	0.7324	Best overall predictive performance; captures complex non-linear degradation.
Random Forest Regressor	23.64	29.92	0.5275	Robust against localized noise and sensor variance.
Decision Tree Regressor	23.81	31.91	0.4625	High interpretability; segment-based rule extraction.
Domain Piecewise Model	27.76	37.53	0.2566	Custom civil engineering traffic-binned regression.
Multi-Linear Regression	35.08	43.04	0.0224	Fails to capture non-linear structural degradation thresholds.

---

📐 Pavement Condition Standards & Treatment Trigger

Predicted roughness values are mapped to Federal Highway Administration (FHWA) and World Bank serviceability criteria:

IRI Range ($\text{in/mi}$)	IRI Range ($\text{m/km}$)	FHWA Condition Rating	Ride Quality	Recommended Highway Maintenance
$< 95$	$< 1.50$	Good	Smooth Ride	Routine maintenance, crack sealing, drainage check
$95 - 170$	$1.50 - 2.70$	Fair	Acceptable	Preventative maintenance (slurry seal / micro-surfacing)
$> 170$	$> 2.70$	Poor	Deteriorated / Rough	Structural rehabilitation (mill & overlay or deep reconstruction)

---

🚀 Quickstart & Interactive Usage

1. Installation

git clone https://github.com/MuhammadUmerBasit/Pavement-Condition-Prediction-ML.git
cd Pavement-Condition-Prediction-ML
pip install -r requirements.txt

2. Predict IRI from the Command Line

Run predict.py with custom pavement parameters:

python src/predict.py --sn 8.5 --aadt 4500

Output:

============================================================
   PAVEMENT CONDITION & ROUGHNESS (IRI) PREDICTION
============================================================
 Input Structural Number (SN) : 8.50
 Input Traffic Volume (AADT)  : 4,500 veh/day
------------------------------------------------------------
 Predicted IRI (in/mi)        : 221.87 in/mi
 Predicted IRI (m/km)         : 3.502 m/km
 Condition Classification     : Poor
 Ride Quality Standard        : Deteriorated / Rough Surface
 Recommended Maintenance      : Structural rehabilitation (mill & overlay or deep reconstruction)
============================================================

3. Run Model Evaluation & Generate Plots

python src/evaluate.py

4. Interactive Jupyter Notebook

Open notebooks/pavement_deterioration_ml.ipynb in VS Code, JupyterLab, or upload to Google Colab for interactive exploratory data analysis and model experimentation.

---

📂 Repository File Structure

Pavement-Condition-Prediction-ML/
│
├── README.md                                  # Project overview and benchmarks
├── LICENSE                                    # MIT License
├── requirements.txt                           # Python dependencies
├── .gitignore                                 # Ignores build artifacts and caches
│
├── dataset/
│   ├── pavement_traffic_dataset.csv          # Clean dataset (75 highway segments)
│   └── raw/                                   # Original field data workbooks
│       ├── Training Data.xlsx
│       ├── SN Data Sangjani Toll Plaza - Attock Bridge.xlsx
│       ├── Traffic Count Sangjani to Attock Bridge.xlsx
│       └── iri rds.xlsx
│
├── src/                                       # Python source package
│   ├── __init__.py
│   ├── model.py                              # Machine learning & domain models
│   ├── evaluate.py                           # Evaluation benchmark script
│   └── predict.py                            # Interactive CLI predictor
│
├── notebooks/                                 # Interactive notebooks
│   └── pavement_deterioration_ml.ipynb       # Jupyter notebook for EDA & training
│
├── models/                                    # Deployment artifacts
│   ├── Machine Learning Model.xlsm           # Interactive Excel VBA deployment
│   └── Sheet1.cls                            # VBA automation script
│
├── docs/                                      # Engineering reports & field documentation
│   ├── Graphical-Representation-IRI-RSL.pdf
│   ├── Pavement-Condition-Roughness-Analysis-Report.pdf
│   ├── Pavement-Structural-Analysis-HWD-Report.pdf
│   └── Annual-Maintenance-Plan-Vehicle-Operating-Costs.pdf
│
└── media/                                     # High-resolution benchmark figures
    ├── model_benchmark_comparison.png
    ├── actual_vs_predicted_iri.png
    └── pavement_condition_breakdown.png

---

👤 Author & Academic Attribution

• Student / Author: Muhammad Umer Basit
  Civil Engineering Student
• Course: Transportation Engineering / Traffic Machine Learning, Semester 8

---

📄 License

This repository is distributed under the MIT License.