The accuracy report indicates the level of accuracy of the IRE. Based on this report, it is possible to investigate which products require the collection of additional information and data to study and improve the tool. In the context of AI and Image Recognition, the Accuracy metric measures how well the AI tool can correctly identify products on a retail shelf from captured images. It measures the percentage of correct identifications made by the AI, taking into account any manual corrections made by users to the initial AI results.
Methodology:
Image Capture: Capture a photo of a retail shelf.
AI Analysis: Allows AI to identify and label the products in the image.
User Correction: Users manually review and correct AI identifications if necessary.
Calculation: Calculate accuracy based on the number of correct identifications by AI and corrections made by users.
Formula: (Number of correct identifications / Total number of identifications made by AI) ×100
Importance: This metric is crucial to continuously improve AI learning and identification capabilities, ensuring more reliable and accurate results over time. It provides clear insights into the performance of the AI tool, allowing for necessary adjustments and improvements.
Usage example: Suppose the AI identifies 4 products of type or label X on the shelf. The user manually corrects the identification, changing one product to type or label Y. Originally, 4 identifications would mean 100% accuracy, but with the correction, only 3 identifications are correct, resulting in an accuracy of 80%. This correction not only impacts the AI’s learning, but also provides the back-office user with an accurate percentage for that period. The accuracy presented to the back-office user is averaged over days and discoveries, allowing for a dynamic and evolving accuracy metric that reflects the AI tool’s learning and improvement journey.
This dedicated page provides a comprehensive analysis of our Machine Learning (ML) system’s accuracy in identifying products. The layout is structured as a list of products, each accompanied by specific columns that detail the ML’s performance for that product. Here’s a breakdown of the columns:
1. Product Name: Displays the specific name of each product.
2. Correct: Indicates whether the ML correctly identified the product. A mark in this column means accurate identification without user intervention.
3. Edited: Shows instances where the application user had to adjust or correct the initial ML identification.
4. New: Highlights cases where the user had to add a new product because ML did not initially recognize it.
5. Deleted: Points out cases where the user removed the ML identification overlay, suggesting that the ML incorrectly identified that product.
6. Accuracy: This represents the percentage of accuracy of ML identification for that specific product. This is demonstrated visually with a vertical bar graph, giving a quick visual representation of ML accuracy for each product. By providing these detailed metrics, this page aims to give users a clear understanding of ML performance, ensuring transparency and aiding in continuous improvement efforts.