🪐 “Decode Dr. Klaus Reinhardt’s final transmission and classify the galaxies to save humanity.”
The Galactic Classification Challenge (GCC) — also known as Cosmic Classifier — is an official ML competition under Cognizance 2025, IIT Roorkee.
Participants must develop an intelligent classifier to predict the type of planets based on their environmental and physical characteristics.
The fate of humankind lies in your hands ⚡
In 2547, Dr. Klaus Reinhardt — a visionary space explorer — transmitted his final dataset moments before his ship was consumed by a black hole.
The data, distorted by gravitational interference, holds the key to identifying habitable worlds.
Your mission: reconstruct, clean, and classify planets using machine learning techniques to ensure humanity’s survival.
You are provided with a dataset containing 10 attributes per planet.
Your goal is to classify each planet into one of 10 German-named categories.
| Feature | Description |
|---|---|
| Atmospheric Density | Thickness of the planet’s atmosphere (kg/m³) |
| Surface Temperature | Average temperature on the planet (Kelvin) |
| Gravity | Surface gravitational acceleration (m/s²) |
| Water Content | Percentage of surface covered by water (0–100 %) |
| Mineral Abundance | Index representing valuable mineral presence (0–1) |
| Orbital Period | Time the planet takes to orbit its star (days) |
| Proximity to Star | Distance from host star (AU) |
| Magnetic Field Strength | Magnetic field intensity (Tesla) |
| Radiation Levels | Average radiation (Sieverts/year) |
| Atmospheric Composition Index | Suitability of atmosphere (0–1 scale) |
- Bewohnbar – Habitable
- Terraformierbar – Terraformable
- Rohstoffreich – Resource-Rich
- Wissenschaftlich – Scientific
- Gasriese – Gas Giant
- Wüstenplanet – Desert World
- Eiswelt – Ice World
- Toxischetmosäre – Toxic Atmosphere
- Hohestrahlung – High Radiation
- Toterahswelt – Dead World
Build a multi-class classification model that predicts the type of a planet using the provided 10 attributes.
The dataset contains noise and missing labels (caused by black-hole interference).
📈 Evaluation Metric: Accuracy
| File | Description |
|---|---|
train.csv |
60 000 samples with labels |
test.csv |
10 000 samples (no labels) |
submission.csv |
Columns → Planet_ID, Predicted_Class |
⚠️ Missing values are represented by large negative numbers — filter them before training.
-
Data Preprocessing
- Handle missing values
- Scale numerical features
- Encode labels
-
Model Training
- Algorithms: Decision Tree, Random Forest, XGBoost, Logistic Regression
- Hyperparameter tuning via RandomizedSearchCV
-
Evaluation
- Compare Accuracy, F1 Score, ROC-AUC
-
Deployment
- Space-themed Streamlit Dashboard (
app.py) - Real-time classification and visualization
- Space-themed Streamlit Dashboard (
| Category | Tools / Libraries |
|---|---|
| Language | Python 3.9 + |
| Frontend | Streamlit + Custom CSS |
| ML Algorithms | Decision Tree, Random Forest, XGBoost, Logistic Regression |
| Libraries | NumPy • Pandas • Scikit-learn • Plotly |
| Visualization | Plotly Express • Seaborn |
| Notebook IDE | Jupyter / Colab |
A beautiful dark space-themed dashboard with real-time prediction and feature analysis.
Features of the App 🌠
- Adjust planetary parameters using sliders
- Visualize feature importance
- See confidence scores & classification results
- Export analysis as CSV
cosmic-classifier/
│
├── app.py # Streamlit Frontend
├── model.ipynb # Model Training Notebook
├── decision_tree_model.pkl # Trained Model
├── x.xlsx # Input Features
├── y.xlsx # Target Labels
├── architecture_diagram.png # Workflow Image
├── images/
│ └── space_background.jpg # Background Image
└── README.md
git clone https://github.com/yourusername/cosmic-classifier.git
cd cosmic-classifier
pip install -r requirements.txt🚀 Run the App
streamlit run app.py
⚖️ License
Licensed under the MIT License. Free for educational and research use.
⭐ If you found this project interesting, don’t forget to star it on GitHub!