edgaracarneiro/feup-ecac

Project developed for 'Knowledge Extraction and Machine Learning', a fifth year subject @FEUP. Made in collaboration with @cyrilico.

feup-ecac's Issues

Attempt messing with class weights on prediction algorithms

Plot DT and RF Algorithms

MAKE PUBLIC PLS

soares is dead, long live ecac

Make K folds random

I have a 1k dataset, assuming a k folda value of 0.25 we do:

• Get 250 of the 1k and run them with the other 750;
• Get 250 of the 750 and run them with the other 750;
• Get 250 of the 500 and run ..
  ...

Run all prediction methods, instead of only one

Instead of forcing the User to choose a prediction method in each run, run all available and output the two models with highest performance and the respective predictions.

For the predictions instead of only using the test files, use the test + train files

Will it get better?

Unir account + district com client, pelas respectivas keys

Experiment with under sampling

Make analysis of the current algorithms, such as DT and RF

Fix undersampling

Remove redundant data (e.g. duration) and use PCA to assert the remaining features

PCA - Principal Component Analysis

Experiment with over-sampling