| _version_ | 1866901607261667328 |
|---|---|
| author | MartinVild |
| author_facet | MartinVild |
| contents | <h1>Steel Joint Prediction using Neural Networks</h1> <p>This repository contains Python scripts for training and evaluating a regression model to predict the plastic resistance of steel joints using features derived from cross-sectional geometry and material properties. The model is built using PyTorch and includes preprocessing, filtering, scaling, training, and visualization.</p> <h2>Contents</h2> <ul> <li><code>LoadNearUnstiffenedEnd.py</code>: Script for generating data using IDEA StatiCa REST API</li> <li><code>ML.py</code>: Main script for training and evaluating the regression model.</li> <li><code>Functions.py</code>: Utility functions for data preprocessing, scaling, plotting, and model construction.</li> <li><code>ML_evaluation.py</code>: Script to predict outputs based on set features using evaluation_model.pth</li> <li><code>Reliability.py</code>: Script to assess reliability based on SAFEBRICTILE procedure</li> <li><code>shap_analysis_model.py</code>: Script for creating shap_analysis.pkl</li> <li><code>shap_analysis_model.py</code>: Script for generating SHAP graphs out of shap_analysis.pkl</li> <li><code>Plots/</code>: Folder containing generated plots during training and evaluation.</li> <li><code>requirements.txt</code>: Python dependencies required to run the scripts.</li> </ul> <h2>Features</h2> <ul> <li>Bulk numerical model calculations to generate a dataset</li> <li>Data filtering based on sparse bins</li> <li>Similitude-based scaling</li> <li>Custom loss function: Relative Squared Error</li> <li>Configurable neural network architecture</li> <li>Early stopping and learning rate scheduling</li> <li>Evaluation metrics: Bias and Coefficient of Variation (CoV)</li> <li>Visualization of training progress and prediction accuracy</li> <li>Reliability assessment using Eurocode procedures</li> <li>Explainability of machine-learned predictions by SHAP</li> </ul> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17359933 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | MartinVild/Machine-Learning: Steel Joint Prediction using Neural Networks MartinVild <h1>Steel Joint Prediction using Neural Networks</h1> <p>This repository contains Python scripts for training and evaluating a regression model to predict the plastic resistance of steel joints using features derived from cross-sectional geometry and material properties. The model is built using PyTorch and includes preprocessing, filtering, scaling, training, and visualization.</p> <h2>Contents</h2> <ul> <li><code>LoadNearUnstiffenedEnd.py</code>: Script for generating data using IDEA StatiCa REST API</li> <li><code>ML.py</code>: Main script for training and evaluating the regression model.</li> <li><code>Functions.py</code>: Utility functions for data preprocessing, scaling, plotting, and model construction.</li> <li><code>ML_evaluation.py</code>: Script to predict outputs based on set features using evaluation_model.pth</li> <li><code>Reliability.py</code>: Script to assess reliability based on SAFEBRICTILE procedure</li> <li><code>shap_analysis_model.py</code>: Script for creating shap_analysis.pkl</li> <li><code>shap_analysis_model.py</code>: Script for generating SHAP graphs out of shap_analysis.pkl</li> <li><code>Plots/</code>: Folder containing generated plots during training and evaluation.</li> <li><code>requirements.txt</code>: Python dependencies required to run the scripts.</li> </ul> <h2>Features</h2> <ul> <li>Bulk numerical model calculations to generate a dataset</li> <li>Data filtering based on sparse bins</li> <li>Similitude-based scaling</li> <li>Custom loss function: Relative Squared Error</li> <li>Configurable neural network architecture</li> <li>Early stopping and learning rate scheduling</li> <li>Evaluation metrics: Bias and Coefficient of Variation (CoV)</li> <li>Visualization of training progress and prediction accuracy</li> <li>Reliability assessment using Eurocode procedures</li> <li>Explainability of machine-learned predictions by SHAP</li> </ul> |
| title | MartinVild/Machine-Learning: Steel Joint Prediction using Neural Networks |
| url | https://doi.org/10.5281/zenodo.17359933 |