A weighted loss function assigns different levels of importance to different classes or samples during model training, which is particularly useful for handling imbalanced datasets. By adjusting the contribution of each class to the overall loss, it helps the model focus more on underrepresented classes, improving predictive performance across all categories.