Cracks to Craters

AI-Based Road Damage Detection & Risk Prediction

Try it out at https://cracks-to-craters.onrender.com/ (cold starts)

Find the python notebook detailing the training of the model in Notebooks\Pothole-formation-risk-model

Overview

Cracks to Craters is a machine learning system designed to detect and predict road damage severity based on environmental and usage factors. The goal is to move from reactive pothole fixing to proactive maintenance.

Features

Predict pothole formation risk
Uses structured input data (rainfall, traffic, pavement age, etc.)
ML-based classification model
FastAPI backend for real-time predictions
Scalable deployment (local + cloud)

Problem Statement

Road damage like potholes develops due to multiple factors:

Traffic load
Rainfall
Pavement age
Soil type

Traditional systems:

Detect damage after it appears
Waste time and resources

This system:

Predicts risk before failure

System Architecture

          ┌──────────────────────┐
          │   Input Features     │
          │----------------------│
          │ Rainfall             │
          │ Traffic Volume       │
          │ Pavement Age         │
          │ Soil Type            │
          └─────────┬────────────┘
                    │
                    ▼
          ┌──────────────────────┐
          │  Feature Processing  │
          │----------------------│
          │ Encoding             │
          │ Scaling              │
          └─────────┬────────────┘
                    │
                    ▼
          ┌──────────────────────┐
          │  ML Model            │
          │----------------------│
          │ Logistic Regression  │
          └─────────┬────────────┘
                    │
                    ▼
          ┌──────────────────────────────┐
          │       Prediction Output      │
          │------------------------------│
          │ Risk Level(Low, Mediu, High) │
          └──────────────────────────────┘

Tech Stack

Python
FastAPI
Scikit-learn
Pandas
NumPy
Pydantic

Dataset

Feature	Description
avg_rainfall_mm	Average rainfall
traffic_volume_vph	Vehicles per hour
pavement_age_yrs	Age of road
last_repair_yrs	Years since last repair
soil_type	Type of soil

Model

Logistic Multivariable Regression

Graphs

Graph 1
Graph 2

Confusion Matrix

API Usage

Endpoint

POST /predict

Request

{
  "avg_rainfall_mm": 120,
  "traffic_volume_vph": 300,
  "pavement_age_yrs": 5,
  "last_repair_yrs": 2,
  "soil_type": "clay"
}

Response

{
  "prediction": "medium"
}

Installation

git clone https://github.com/fourtysevencode/cracks-to-craters.git  
cd cracks-to-craters  
pip install -r requirements.txt

Run

uvicorn app:app --reload

Authors

Ronak Pal, Osh Singh, Sri Katyayani Kota