GitHub Repositories:

- https://github.com/DenisPablo/ExoPlanetas

- https://github.com/Hyouka21/ExoPlanetas-InterSoft-SL-Frontend.git

Introduction

The Intersoft project consists of the development of an intelligent model based on Artificial Intelligence (AI) designed for the automatic detection of exoplanets outside our solar system. Using the open data from the Kepler mission (KOI Dataset) and by applying the transit method, the system analyzes variations in stellar brightness to determine the presence of possible planets.

The purpose of the project is to automate and optimize the process of identifying exoplanets, reducing manual intervention and improving efficiency in the classification of astronomical candidates, thereby supporting future scientific research.

The project was developed by the Intersoft team, composed of an interdisciplinary group of students and enthusiasts of technology and astronomy.

The development environment was local, but the application was later hosted in the cloud, enabling remote access and scalability for future deployments.

Exploring the ExoClassifier

A Brief Tour of Our App

Dashboard:

Classify:

Upload CSV:

Model Training:

Model Information:

Conclution

The Intersoft project aims to become a reference tool for the automated detection of exoplanets. Its scalable structure allows for the incorporation of new features. The system will be publicly accessible, available on the web to any user interested in research or space observation.

The Intersoft project represents a combination of science, technology, and teamwork. The team succeeded in designing an AI model that is efficient, accurate, and visually interpretable, applying advanced machine learning concepts to real astronomical data. During development, the team strengthened their ability to collaborate, manage time, and solve complex problems, gaining comprehensive experience across the entire lifecycle of creating a scientific-technological project.

In the team’s words: “An exciting, challenging, and ambitious project that tested us, kept us constantly learning, and, without a doubt, was an unforgettable experience.”

Technical Details

Technologies used

- Languages and frameworks: Python, PHP, JavaScript, React.

- AI and data analysis libraries: TensorFlow, Keras, Pandas, NumPy.

- Machine Learning techniques: Random Forest and also we made a neural network.

Methodology

The work process included the following stages:

1. Dataset selection and preparation (KOI):

- Key variables were chosen such as orbital period, total transit duration, transit depth, planet radius, stellar radius, and stellar effective temperature.

- Irrelevant columns were removed and normalization techniques were applied to optimize the statistical consistency of the data.

2. Data splitting:

The dataset was split 80% for training and 20% for testing, seeking a balance between learning and validation.

3. Training and classification:

The model classifies objects into three categories:

- Confirmed

- Candidate

- False positive

In addition, it provides a probability percentage associated with each prediction, offering an interpretable confidence level for the user.

4. Performance evaluation:

Metrics such as confusion matrix, accuracy, and F1-score were used to determine performance.

Models can be retrained dynamically, allowing parameters to be adjusted and accuracy to be improved with new data.

Interface and operation

The system features an interactive, responsive web interface developed in React and connected to the models via API services. The application allows the user to:

- Upload new data or use existing datasets.

- Graphically visualize classification results.

- Consult model performance statistics.

- Adjust parameters to retrain the system.

The design aims for the tool to be accessible, intuitive, and useful for both scientists and students in the astronomical community, promoting collaboration and scientific exploration.

Behind the Scenes

During development, significant challenges arose:

- Noise in the data, which caused fluctuations in the results.

- Class imbalance, which made statistical normalization difficult.

These issues were addressed through variable filtering and the implementation of normalization techniques and dynamic parameter tuning.

upload_image.21:32:46.178014 upload_image.21:33:12.857880

2025 NASA Space Apps Challenge

Intersoft Team