Inspiration
Space has always fascinated me. I have done science Olympiads and national science bowls, where my main events were always related to space. I wanted to use technology to explore space in a more interactive and insightful way. I was inspired by the opportunity to combine my coding skills with my curiosity about the universe. The Stellar Gateway Hackathon seemed like the perfect chance to push myself to learn more about space, data, and machine learning all at once.
What it does
AstroPredict is a powerful AI tool that lets you predict solar flares and visualize space data in real time. Using historical solar flare data and machine learning, it forecasts future solar events, helping you:
Analyze past solar activity: View patterns and trends in solar flares over time.
Compare different models: See how Logistic Regression, SVM, and Random Forest perform in real-world conditions.
Make predictions: Input a set of conditions and let the trained models predict whether a solar flare is likely to happen soon.
Visualize results: Display all this information in interactive charts and graphs, making space data more accessible and easier to explore.
How we built it
I started by gathering solar flare data from a NASA satellite mission (GOES). Using Python’s ecosystem; pandas for cleaning, scikit-learn for modeling, and matplotlib for visualization. I prepared the data and trained multiple models (like Logistic Regression and Random Forest) to predict future solar flares. Then I integrated everything into a Streamlit application. This lets users select a model, view its performance, and visualize solar flare patterns, all through a simple, interactive GUI.
Challenges we ran into
Some key challenges I faced were:
Messy and incomplete data: The raw solar data was unsorted and had missing values. I had to learn techniques to clean, fill in, and transform it to be useful for training.
Model selection: There were many different algorithms I could use. I experimented with a few to find which performed well for this specific problem.
Designing a clear UI: Making sure the Streamlit app was simple, informative, and easy to navigate was challenging, especially when I was trying to show a lot of information.
Balancing complexity and interpretability: I wanted accurate models, but I also wanted them to be interpretable and lightweight enough to run smoothly in a Streamlit application.
Accomplishments that we're proud of
I successfully cleaned messy solar flare data and trained a powerful ML pipeline that can make accurate predictions about future solar activity.
From data collection and cleaning, to model training, to UI development with Streamlit, I built a fully functional tool on my own; not just a piece of code, but an interactive application.
Through this project, I learned how to handle large datasets, apply machine learning algorithms, and create a polished UI, all in a short amount of time. I’m proud of having gone from a beginner to developing a sophisticated tool.
What we learned
Through this project, I learned how to handle large, messy datasets and clean them for analysis, and I explored different machine learning models to find which ones best predicted solar flares. I also got hands-on experience designing a Streamlit application, making sure it was interactive, clear, and useful to both space enthusiasts and data practitioners. This journey challenged me to connect coding, data science, and UI design, and it deepened my understanding of how we can use technology to solve real-world problems.
What's next for AstroPredict
Add More Data: I plan to incorporate new datasets, including sunspot numbers, solar wind speeds, and magnetic field measurements, to make the predictions more accurate.
Improve Model Accuracy: I want to experiment with deep neural nets or time-series models (like LSTM) to better capture patterns in solar activity over time.
Real-Time Predictions: My future aim is to connect AstroPredict directly to real-time satellite feeds, allowing it to provide up-to-date forecasts and alerts.
Mobile App: I’d like to develop a mobile-friendly or phone application to make this tool more accessible to space enthusiasts, educators, and journalists.
Public API: I'll consider adding a API endpoint that lets other developers retrieve solar flare predictions and data for their own applications.
Log in or sign up for Devpost to join the conversation.