Saved in:
| Main Authors: | , |
|---|---|
| Format: | Recurso digital |
| Language: | Portuguese |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.17281701 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>This repository accompanies the Master’s dissertation <em>“Development of flexible pavement performance prediction models for the states of RS and PR using artificial intelligence”</em> (Camillo, 2025). It contains the full implementation of the predictive modeling framework used to estimate short-term pavement condition indicators based on historical monitoring data.</p> <p>The system is implemented in <strong>Python</strong> and employs <strong>XGBoost</strong> models previously trained for each pavement performance indicator — including <em>IRI</em>, <em>PSI</em>, <em>D0</em>, <em>ATRMED</em>, and <em>IDS</em>. The inference pipeline automates the prediction process across multiple pavement ages by:</p> <ol> <li> <p>Filtering and preprocessing the dataset by year;</p> </li> <li> <p>Loading the appropriate trained model and feature scaler (<code>joblib</code> + <code>.json</code>);</p> </li> <li> <p>Scaling input features and performing inference for the next period;</p> </li> <li> <p>Applying logic to prevent unrealistic improvements or deterioration without maintenance events (reinforcement or cut);</p> </li> <li> <p>Sequentially updating the dataset and exporting the predicted evolution of each indicator to Excel;</p> </li> <li> <p>Generating <strong>trend plots</strong> for each group or highway, visually representing deterioration over time.</p> </li> </ol> <p>The methodology follows the four-step structure proposed in the dissertation:</p> <ul> <li> <p><strong>Step 1:</strong> Data exploration, cleaning, and feature selection;</p> </li> <li> <p><strong>Step 2:</strong> Algorithm comparison and hyperparameter optimization (PSO and Grid Search);</p> </li> <li> <p><strong>Step 3:</strong> Evaluation of PCA and oversampling preprocessing methods;</p> </li> <li> <p><strong>Step 4:</strong> Validation and interpretation of model outputs through both quantitative and qualitative criteria.</p> </li> </ul> <p>The resulting system demonstrates the feasibility of combining engineering domain knowledge with artificial intelligence to model pavement deterioration in Brazilian highway conditions, offering a reproducible and extensible framework for performance prediction in pavement management systems.</p> <p><strong>Technologies used:</strong> Python, XGBoost, NumPy, Pandas, Joblib, Matplotlib.<br><strong>Main outputs:</strong> <code>_Dados sinteticos - Processado.xlsx</code> and time-evolution charts per indicator.</p> <p><strong>Keywords:</strong> Pavement performance, Artificial intelligence, Machine learning, XGBoost, Predictive modeling, Pavement management, Brazil.</p>