We will also try to classify the dataset using XGBoost. As XGBoost uses trees of a maximum depth of three, we expect that it will outperform AdaBoost without any fine-tuning. Indeed, XGBoost is able to achieve better performance in both datasets and for all metrics (as shown in the following table), compared to most previous ensembles:
|
Dataset |
Metric |
Value |
|
Original |
F1 |
0.846 |
|
Recall |
0.787 |
|
|
Filtered |
F1 |
0.849 |
|
Recall |
0.809 |
XGBoost out-of-the-box performance
By increasing the maximum depth of each tree to five, the ensemble is able to perform even better, yielding the following results:
|
Dataset |
Metric |
Value |
|
Original |
F1 |
0.862 |
|
Recall |
0.801 |
|
|
Filtered |
F1 |
0.862 |
|
Recall |
0.824 |
Performance with max_depth=5