SVM Weights

SVM models are automatically initialized to achieve the best average prediction across all classes. If the training data does not represent a realistic distribution, you can bias the model to compensate for class values that are under-represented. If you increase the weight for a class, the percent of correct predictions for that class should increase.