Explainable AI: Machine Learning Interpretation in Blackcurrant Powders.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Explainability in machine and deep learning is crucial due to the increasing use of AI.
• Explainable AI (XAI) improves transparency and effectiveness of AI model decisions.
• XAI aids in data mining, error elimination, and enhancing AI algorithm performance.

Purpose of the Study:

• To understand the identification of selected currant powder types using 'glass box' and 'black box' AI models.
• To evaluate the performance of AI models in classifying currant powders based on texture descriptors.
• To visualize model explanations using Local Interpretable Model Agnostic Explanations (LIMEs).

Main Methods:

• Utilized Decision Tree and Random Forest models for currant powder identification.
• Trained models using texture descriptors: entropy, contrast, correlation, dissimilarity, and homogeneity.
• Assessed model performance using accuracy, precision, recall, and F1-score metrics.
• Employed Local Interpretable Model Agnostic Explanations (LIMEs) for visualization.

Main Results:

• Bagging (Bagging_100), Decision Tree (DT0), and Random Forest (RF7_gini) were the most effective models.
• Bagging_100 achieved approximately 0.979 for accuracy, precision, recall, and F1-score.
• DT0 and RF7_gini models demonstrated classifier performance measures exceeding 96%.

Conclusions:

• XAI models, particularly Bagging, Decision Tree, and Random Forest, are effective for identifying currant powders.
• The study highlights the potential of XAI in analyzing food product data.
• Agnostic XAI models can serve as valuable tools for online data analysis in the future.