A binary classifier based on multivariate Gaussian models, which estimates the mean vector and the variance-covariance matrix during the training phase and returns the class with the highest likelihood during the test phase, can generally be considered more resistant to the problem of an imbalanced training set compared to other classifiers (e.g., discriminative ones)?
Isuru Senarath
No, a binary classifier based on Gaussian models like Gaussian Naive Bayes is generally not robust to severe training set imbalance. It tends to be biased towards the majority class and may perform poorly on predicting the minority class in highly imbalanced datasets.
2 votes
1 thanks
- Badges
- Science topic
- Similar topics
- Comment
More Francesco Sigona's questions See All
Similar questions and discussions