TrainingScopes

Use Normal training scope to include every page of each training sample in the classification model.

• Best for documents where identifying features may appear on any page.
• Increases the amount of training data, which can improve accuracy but may also introduce noise if middle pages are not distinctive.

Recommended for unstructured documents with variable or distributed features.

Use FirstLast training scope to include only the first and last pages of each training sample.

• Ideal for documents where the first and last pages contain the most distinctive features (e.g., headers, salutations, signatures).
• Excludes middle pages, reducing noise and improving performance for documents with repetitive or generic content in the middle.

Recommended for correspondence, letters, or contracts with clear start and end markers.

Use FirstOnly training scope to include only the first page of each training sample.

• Best for documents where all identifying features are present on the first page.
• Minimizes training data and maximizes performance, but may reduce accuracy if features appear elsewhere.

Recommended for forms, cover sheets, or documents with unique first-page layouts.

Use FML (First, Middle, Last) training scope to train separate models for first, middle, and last pages.

• Enables advanced classification by distinguishing between different page roles within a document.
• Improves accuracy for documents where the first, middle, and last pages have distinct features or layouts.
• Requires more training samples to fully represent each page class.

Recommended for complex unstructured documents with variable-length bodies and distinctive start/end pages.