Word
feature attribution
Definition
n. a technique that calculates which input words, pixels, or facts were most responsible for an AI model's final decision.
n. methods that assign numerical importance scores or credit to individual input features based on their contribution to a model's specific prediction.
Examples
“Feature attribution showed that the credit algorithm rejected the loan primarily due to the applicant's debt ratio.”
“Doctors use feature attribution heatmaps to see which regions of an X-ray triggered the pneumonia alert.”
“Integrated Gradients guarantees axiomatically sound feature attribution by integrating path gradients from a neutral baseline to the target input.”
Examples
simple
“Feature attribution showed that the credit algorithm rejected the loan primarily due to the applicant's debt ratio.”
contextual
“Doctors use feature attribution heatmaps to see which regions of an X-ray triggered the pneumonia alert.”
complex
“Integrated Gradients guarantees axiomatically sound feature attribution by integrating path gradients from a neutral baseline to the target input.”
Real-World Examples
“One of the most influential directions within XAI is feature attribution, which assigns an importance score to each input feature for a specific prediction [2].” “Feature attributions based on the Shapley value are popular for explaining machine learning models.” Etymology
Compound machine learning term combining feature (from Latin factura, 'formation, make') with attribution (from Latin attribuere, 'to assign, credit').
Etymology adapted from Wiktionary, available under CC BY-SA 4.0.