Imbalanced dataset with 3 classes xgboost scale_pos_weight parameter

Question

alexprice

2022年5月5日 08:09

The xgboost classifier states the use of parameter scale_pos_weight for 2-class problems.

I have a highly imbalanced dataset with 3 classes. Classes '1' and '-1' are very rare (~1% of dataset) and class '0' is very common.

How do I set this scale_pos_weight parameter in the xgboost classifier correctly for my classification problem?

Bing Lv · Accepted Answer · 2022年5月5日 08:09

The parameter scale_pos_weight works for two classes (binary classification).

The parameter weight goes into the xgb.DMatrix function can be used for three or more classes. The weights can be computed like this:

weights = total_samples / (n_classes * class_samples * 1.0)

wieus · Accepted Answer · 2019年1月8日 08:42

For my multiclass classification problem with similar unbalanced data I used the output from sklearn compute_class_weight function:

sklearn.utils.class_weight.compute_class_weight(class_weight, classes, y)