DanielEmeka/FormationEnergyPredictor

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

📘 Formation Energy Predictor

FormationEnergyPredictor is a machine learning web application that predicts the formation energy (eV/atom) of perovskite-based materials using their atomic and physical properties. It also classifies the material as Stable, Metastable, or Unstable, helping materials scientists rapidly screen novel compositions for energy, electronics, and photovoltaic applications.

🌐 Live App on Hugging Face Spaces

🔬 What It Does

• —Takes in a material’s A site, B site, and X site elements and a few physical properties
• —Predicts the formation energy using a trained machine learning model
• —Classifies material stability:

• —🟢 Stable (low/negative formation energy)
• —🟡 Metastable (near-zero formation energy)
• —🔴 Unstable (positive formation energy)
• —Visualizes the result on a stability scale

🏗️ How It Was Built

This project was built using:

• —Python & Scikit-learn: for training a Gradient Boosting model
• —Gradio: for building the web interface
• —Matplotlib: to visualize formation energy in a stability chart
• —Hugging Face Spaces: to host the application online

🔧 Workflow

1. 1.Data Source: A curated dataset of perovskite materials with elemental properties and formation energies.
2. 2.Feature Engineering:

• —Selected top features including:

• —A site #1, B site #1, X site
• —Number of elements
• —Density_AB_avg
• —Ionization Energy_AB_avg
• —Atomic Volume_AB_avg
• —Modeling:

• —Used GradientBoostingRegressor inside a preprocessing pipeline
• —Trained with 80/20 train-test split and achieved a good R² score
• —Deployment:

• —Packaged with gradio and deployed to Hugging Face using requirements.txt

🧠 Use Case & Problem Solved

🚧 Challenge:

In materials science, discovering new stable compounds is time-consuming and expensive. Researchers often simulate or synthesize hundreds of compositions before finding one that works.

✅ This app solves:

• —Fast pre-screening of candidate perovskite materials
• —Prediction of stability before physical experiments or simulations
• —Reduction in cost and time spent on unstable materials
• —Supports green energy and optoelectronic research

▶️ How to Use

1. 1.Visit the app: FormationEnergyPredictor on Hugging Face
2. 2.Fill in:

• —A site element (e.g., Na)
• —B site element (e.g., Ti)
• —X site element (e.g., O)
• —Additional properties like density, ionization energy, etc.
• —Click Submit
• —View:

• —Formation energy prediction (eV/atom)
• —Stability classification
• —Visual chart showing result

💡 Examples to Try

📌 Key Feature Insights

From the trained model, these features were found most important:

• —Ionization Energy — higher values lead to instability
• —Density and Atomic Volume — influence bonding compactness
• —A, B, X site identity — define lattice behavior in perovskites

🔮 Future Improvements

• —Add support for ABX₂ and double perovskite structures
• —Enable batch predictions via file upload
• —Integrate SHAP or permutation importance plots for full explainability
• —Add default dropdown menus and example presets

🧑‍🔬 Author

Daniel Emeka Physicist | Nanomaterials Researcher | Machine Learning Enthusiast Built for material discovery and sustainability-driven innovation.