# Exoplanet Settlement Viability Challenge

## Competition Overview

Welcome to the Exoplanet Settlement Viability Challenge, a machine learning competition hosted as an assignment for two courses at [Washington University in St. Louis](https://wustl.edu/). This competition is designed for students enrolled in instructor [Jeff Heaton's](https://sites.wustl.edu/jeffheaton/) courses:

- [T81-558: Applications of Deep Neural Networks](https://github.com/jeffheaton/app_deep_learning)
- [T81-559: Applications of Generative AI](https://github.com/jeffheaton/app_generative_ai)

While primarily intended for these courses, this competition is also open to the broader Kaggle community.

## The Challenge

You are tasked with predicting the **settlement viability index** for newly discovered exoplanets evaluated by an interstellar exploration agency. This comprehensive metric (ranging from 0 to 100) estimates the overall feasibility of establishing a successful, self-sustaining human settlement on each planet. Higher scores indicate planets that are more suitable for settlement.

The settlement viability index synthesizes multiple critical factors:

- **Environmental Habitability (40%)**: Atmospheric conditions, temperature, gravity, oxygen levels, day length, and temperature stability
- **Environment Type Classification (25%)**: The planet's dominant environmental characteristics (temperate, oceanic, desert, volcanic, toxic, etc.)
- **Resource Availability (15%)**: Presence of rare minerals, energy harvest potential, and water availability
- **Safety and Risk Factors (10%)**: Radiation levels, seismic activity, storm frequency, biological threats, and magnetic field strength
- **Economic and Strategic Factors (10%)**: Settlement costs, terraforming difficulty, and strategic value
- **Aesthetic Factors (5%)**: Visual characteristics that affect settler psychological well-being, including albedo and cloud coverage

## Dataset Description

This dataset provides a rich, multi-modal collection of information about each exoplanet, including **structured tabular data**, **unstructured textual data**, and **generated images**.

### Structured Data

The dataset includes extensive scientific measurements and categorical attributes for each planet:

- **Orbital and Physical Properties**: Star type, orbital distance, orbital period, day length, axial tilt, gravity, escape velocity
- **Atmospheric Composition**: Pressure, primary gases, oxygen/CO2/methane percentages
- **Surface Conditions**: Temperature (mean and range), water presence, surface composition, dominant biome
- **Environmental Factors**: Albedo, cloud coverage, storm frequency, seismic activity, magnetic field strength, radiation levels
- **Biological and Resource Data**: Native biology presence, invasive risk, rare mineral index, energy harvest potential
- **Settlement Metrics**: Terraforming difficulty, settlement cost index, strategic value rating, primary hazards, viability classification

### Unstructured Text Data

Each planet includes three narrative text fields that provide qualitative insights:

- **Exploration Log** (200-400 words): First-landing field notes mixing scientific observations with mission-relevant details
- **Environmental Report** (120-250 words): Technical and concise environmental assessment
- **Incident Report** (60-150 words): Documentation of anomalies, equipment issues, biological encounters, or unexplained phenomena

These narratives vary in tone from optimistic to cautious to ominous, reflecting the diverse nature of planetary exploration.

### Image Data

AI-generated images are provided to visualize each exoplanet's surface. These cinematic, wide-angle exterior views depict the landscape features, atmospheric conditions, sky colors, stellar appearance, and environmental effects consistent with each planet's attributes. **Note that not every planet in the dataset may have an associated image.**

![Example Exoplanet Surface](https://s3.us-east-1.amazonaws.com/data.heatonresearch.com/images/kaggle/planets-58.jpg)

_Example: A tidally locked world with perpetual twilight, featuring crystalline formations and a red dwarf star on the horizon_

## Approach Strategies

This competition offers multiple pathways to success, depending on your background and course focus:

**For T81-558 (Deep Neural Networks) students**: You may choose to focus on the structured tabular features combined with computer vision techniques applied to the planetary images. Neural networks can learn complex non-linear relationships between physical properties and settlement viability.

**For T81-559 (Generative AI) students**: You may choose to leverage the unstructured textual data using natural language processing and large language models. The exploration logs, environmental reports, and incident reports contain valuable qualitative insights that can be extracted through modern NLP techniques.

**For the strongest solutions**: Hybrid approaches that combine structured data, image analysis, and text mining will likely achieve the best performance. Multi-modal machine learning techniques that can integrate information across all three data types represent the cutting edge of this challenge.

## Data Diversity

The dataset encompasses a wide variety of planetary conditions to test the robustness of your models:

- Different star types (red dwarfs, G-type main sequence, K-type stars, binary systems)
- Extreme environments (tidally locked worlds, high-pressure ocean worlds, low-gravity relic worlds, methane-sea moons)
- Varied atmospheric compositions and surface conditions
- Edge cases that challenge conventional habitability assumptions

Each planet maintains internal consistency between its attributes (for example, oxygen levels align with biological presence, and temperature aligns with water state).

Good luck, and may your models identify the next home for humanity among the stars!