You trained an SVM for binary classification and got validation AUC=0.87. You want to increase AUC. What should you do?

Performing hyperparameter tuning is the most effective next step to improve the AUC of an existing SVM model. Hyperparameters control the learning process (e.g., C, gamma for SVMs) and finding optimal values can significantly boost model performance. Training a deep neural network isn't guaranteed to outperform an SVM; model choice depends on data characteristics, and SVMs can be highly effective. Deploying the model and expecting higher real-world AUC is incorrect; validation AUC is an estimate, and real-world performance can vary, often being lower due to data drift or distribution shifts. Scaling model outputs by a multiplicative factor won't change the AUC, as AUC is rank-order dependent and scaling merely shifts the decision threshold without altering the relative order of predictions.