An ML specialist builds a credit scoring model using 3 years of transaction and third-party metadata but finds low accuracy on both training and test sets. Which actions would improve model accuracy? (Choose two.)

Low accuracy on both training and test sets indicates high bias (underfitting). Increasing training epochs/passes allows the model more opportunities to learn patterns from the existing data, potentially reducing bias. Performing more hyperparameter tuning helps find the optimal model configuration that can better capture the underlying relationships in the data. Adding new domain-specific features can provide the model with more relevant information, improving its ability to make accurate predictions. Switching to more complex model architectures (e.g., from linear models to neural networks) can enable the model to learn more intricate, non-linear relationships, addressing underfitting. Increasing regularization would further constrain the model, potentially increasing bias. Reducing the number of feature combinations or decreasing numeric attribute binning might simplify the model, exacerbating underfitting. Reducing training data or passes would also likely worsen underfitting.