Asteroid diameter prediction using machine learning models - case study on a kaggle dataset.
This repository holds our 2022 Artificial Intelligence class project, which managed to earn us 3rd place in an Amazon Learn & Earn competition.
We have managed to study and classify 7 distinct ML models, which were trained on a NASA dataset of known asteroids. Using 88 selected features such as orbit parameters and captured image properties, we have trained the following models:
- MLP-NN
- C-NN
- SVM
- AdaBoost
- K-NN (uniform)
- K-NN (distance)
- Random Forest
After analyzing the results based on different metrics, it was clear that the Random Forest model had the best immediate results. However, we are firmly convinced that with careful hyperparameter tuning, we can achieve better results using a Neural Network model.
For a digest of our results, refer to our Learn & Earn presentation (available only in romanian at time of this commit).