Catboost vs RF for goat behavior detection

NOTE: all the experiments were using the six sensors.

Three scripts were used for generating the results:

1. caprino_catboost.R

2. caprino_rf.R

3. caprino_catboost_caret.R

RF vs. catboost

Evaluation using a low iteration number

Checking results for trainset

The iteration number of catboost is critical to avoid overfitting. Results with a very low iteration number are shown below for training dataset.

catboost (various approaches)

Two different approaches for catboost. 1. catboost.cv to estimate the iteration number. 2. catboost using the caret interface and selecting the best model according to CV 3. Random Forest