DOCS

Confidence score

/

Classify confidence score

Learn how Classify's confidence scoring works.

Discover how Zonos Classify utilizes entropy confidence scoring and how to adjust your minimum confidence score.

How it works 

The confidence score in Classify is a numerical value between 0% and 100%, expressed as a decimal, meant to represent a classification's level of certainty. This score is determined using entropy, a method from information theory that measures uncertainty or randomness in the data. As entropy quantifies uncertainty, higher uncertainty leads to lower confidence scores, while lower uncertainty yields higher confidence scores. This approach provides a more nuanced understanding of uncertainty in classifications, resulting in a more accurate confidence score compared to the standard probability approach, which simply quantifies the most likely option.

The confidence score defined here is not a calibrated one. This means that the scores themselves cannot be interpreted in absolute terms. For instance, an 85% score does not necessarily mean that the classification will result in something favorable 85% of the time for your purposes. Instead, these scores are most useful when compared against each other. An 85% confidence classification carries a higher likelihood of favorable outcomes compared to a 75% confidence score.

Consider this scenario: providing only "sweater" or "food" as information might mistakenly suggest both categories have an equal chance of being correct. However, "sweaters" have 10 subheadings, while "other food preparations" encompass 600 subheadings. Entropy-based confidence scores factor in these differences, leading to much higher confidence in the "sweater" classification than in the "food" classification.

Confidence ranges

Available in Dashboard, Classify confidence ranges provide a straightforward and intuitive framework for interpreting confidence scores. Expert evaluation of entropy scoring and data informed the creation of these ranges, aiming to effectively communicate our level of confidence in each classification. This system is designed to minimize ambiguity, making it easier for users to make informed decisions based on the confidence scores.

  • 75-100% = High confidence
  • 50-74% = Moderate confidence
  • 25-49% = Fair confidence
  • 10-24% = Low confidence
  • 0-9% = No result

Adjust minimum confidence score 

While Classify has a default confidence score, you can adjust the score to suit your business needs. If the confidence score is less than your desired percentage, Classify will not return an HS code. Instead, it will notify you an HS code could not be found and prompt you to add more product information.

Classify's confidence score is an essential part of Classify as it ensures we return HS codes within your risk tolerance. For example, if you set your confidence score at 25%, Classify likely will not provide classifications for ambiguous descriptions like "handmade accessories" or "dragon pack." However, by reducing the confidence score to 0%, Classify will make assumptions, such as classifying "handmade accessories" as "bracelets" and "dragon pack" as a "toy", to ensure you receive a classification response whenever possible.

To consult with a classification expert and to adjust your confidence score level, reach out to classify@zonos.com.

Provided HS codes 

If you supply a provided HS code as a Classify input field, we treat that as a known attribute. As such, all classifications will be within that subset of the taxonomy, and the provided confidence score will be limited to that same taxonomy subset. For example, if you provide 62 as a provided HS code for a universal request, the confidence score will be based on digits 3 through 6, as we assume the provided HS code is correct.

Was this page helpful?