Skip to content

Understanding Confidence Scores

Introduction

When extracting data from technical drawings, accuracy is a key factor in ensuring reliable results. The Werk24 API provides confidence scores for extracted information, allowing users to assess the certainty of each result. This article explains how confidence scores work, how they should be interpreted, and best practices for using them effectively.

What Is a Confidence Score?

A confidence score represents the API’s estimation of how certain it is that a given extraction is correct. It is expressed as a value between 0.0 and 1.0, where:

  • 1.0 indicates high certainty (the API is very confident in the extraction).
  • 0.0 indicates no confidence (the API is uncertain about the extraction).
  • Values between 0.0 and 1.0 represent varying degrees of certainty.

Confidence scores are useful when deciding whether extracted data should be used as-is, manually reviewed, or verified against other sources.

Where Are Confidence Scores Used?

Confidence scores are included in most API responses where automatic extraction is performed. This includes:

  • Dimensions: The certainty that a dimension was correctly extracted.
  • Threads: The likelihood that the thread information is correct.
  • Bores, Chamfers, and Radii: Confidence in recognizing these geometric features.
  • Roughness and GD&T Symbols: Certainty that the extracted tolerances and surface roughness data are accurate.

Each extracted entity includes a confidence field within the response payload.

How to Interpret Confidence Scores

Confidence Score Interpretation Suggested Action
≥ 0.95 Very High Confidence Use directly, manual review is rarely needed.
0.85 - 0.94 High Confidence Likely correct, but verify for critical applications.
0.70 - 0.84 Moderate Confidence Review manually before use, especially in important applications.
< 0.70 Low Confidence Manual validation recommended, as errors may be present.

For mission-critical applications (e.g., aerospace, automotive manufacturing), it’s advisable to review values below 0.90 before integrating them into workflows.

Best Practices for Using Confidence Scores

  1. Set a Threshold for Automation: If integrating Werk24 into an automated workflow, define a confidence threshold (e.g., 0.90).Automatically accept values above this threshold and flag lower-confidence results for review.

  2. Choose a Threashold that matches your application: In safety-critical industries, require a higher confidence threshold. For general engineering applications, a lower threshold may be acceptable.

  3. Use Confidence to Prioritize Review: Focus manual verification on results with lower confidence scores first. If human review capacity is limited, start with values below 0.85.