visitor@cyantusk.neocities.org:~$ loadinfo --project ensemble
Loading page information from: Ensemble... Done!
Ensemble Learning
Group: João Gubert and William Weber Berrutti
Technologies used: Python, Linux
Status: Finished
Description
It's the first work (of two) developed for the college's discipline Machine Learning, consisting of implementing ensemble learning using random forests from the ground up.
Problem: The model was learning well even with a single tree, which was unexpected. It should have many mistakes at the start and later fewer.
Challenges and Learnings
- Learning process: We had to use K-Fold with Stratified Bootstrap algorithms. It was quite challenging to understand how it'd function, but we got it working.
- Confusion Matrix: it's a matrix that helps on counting for True Positives, True Negatives, False Positives, and False Negatives. It was hard to think of a generic confusion matrix with all its needed calculations for N target attributes.
- Datasets: we would like to have used more datasets to see how the program would behave.