Machine Learning Analysis

Gentileschi Attribution Study

Dual-Model Computational Attribution System
Analysis completed 2025-09-13 22:47:41

Overview

Total Works
60
Agreement Rate
63.3%
both models same class
High Reattributions
3
> 75% & agreement
Avg Confidence
76.1%
median 74.6%
Date Range
1601–1650
median: 1626
38
Models Agree
3
Reattributions
76%
Avg Confidence
22
Disagreements

Descriptive Statistics

Agreement: 38/60 (63.3%)
Mean confidence: 76.1% · Median: 74.6%
By current attribution: Orazio: 25 · Artemisia: 35
By predicted artist: Artemisia: 44 · Orazio: 16
How to read this
  • Agreement rate = share of works where artist and gender models point to the same side (Artemisia↔Female, Orazio↔Male).
  • Confidence = average of both models' top probabilities per work. Median resists outliers.
  • Cross-tab (below) shows where current catalogue attributions align—or conflict—with model predictions.
  • Use High Reattributions as your priority queue for connoisseurship checks.
Current \ Predicted ArtemisiaOrazio
Orazio1411
Artemisia305

High-Confidence Reattribution Candidates

>75% confidence and model agreement for an attribution different from the catalogue.

Conflicting Model Predictions

Artist-specific vs gender model disagree—possible workshop collaboration or transition.

Models in Agreement

Both attribution models point to the same artist—higher reliability for these predictions.

Confirmed Attributions

>75% confidence supporting the current attribution.

Model Performance Analysis

Evaluation metrics and training history for both attribution models

Artist-Specific Model

81.2%

Validation Accuracy

Confusion Matrix

Artist confusion matrix

Training History

Artist training history

Gender Pattern Model

66.9%

Validation Accuracy

Confusion Matrix

Gender confusion matrix

Training History

Gender training history

Reading the metrics

81.2% Artist Accuracy: separation of Artemisia vs Orazio.

66.9% Gender Accuracy: overlap between male/female stylistic signals.

Model Agreement: highest weight when both models concur at high confidence.