Enterprise AI Analysis
Explainable machine learning-based classification of traditional Korean ceramics using XRF chemical composition data
This analysis details an explainable machine learning framework applied to classify traditional Korean ceramics—celadon, buncheong, and white porcelain—based on XRF chemical composition. Leveraging a curated dataset of 624 samples, our models achieved up to 95.8% classification accuracy. Critical insights were derived on material characteristics, enhancing traditional typological methods with a quantitative, data-driven framework.
Executive Impact: Key Findings at a Glance
Our explainable AI approach delivers robust performance and actionable insights for cultural heritage classification, offering a blueprint for broader enterprise applications in materials science and beyond.
0
Classification Accuracy (RF & XGB)
0
Highest Precision Achieved
0
Highest Recall Achieved
0
Samples Analyzed
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The study rigorously evaluated six machine learning algorithms for classifying Korean ceramics. Tree-based models, specifically Random Forest (RF) and Extreme Gradient Boosting (XGB), achieved the highest classification accuracy of 95.8% on the test set. Other models like SVM also performed strongly at 93.3%. A key finding was the consistent and high accuracy in identifying white porcelain, while celadon and buncheong showed partial misclassification due to their overlapping chemical characteristics. McNemar's test confirmed the statistical significance of the superior performance of ensemble models over simpler linear models like PCA-LDA.
Through Shapley Additive Explanations (SHAP), the analysis identified Fe2O3 and TiO2 as the most influential chemical components for differentiating ceramic types. This aligns well with established ceramic coloration mechanisms. Low concentrations of these oxides were critical for white porcelain, while intermediate to high levels contributed to the distinct hues of celadon and buncheong. PCA scatter plots visually confirmed that white porcelain forms a distinct cluster, whereas celadon and buncheong compositions substantially overlap, reflecting their material and technological continuities.
A curated dataset of 624 traditional Korean ceramic samples (201 celadon, 212 buncheong, 211 white porcelain) was compiled from peer-reviewed XRF chemical composition data (10 major elemental oxides). Data preprocessing included filtering, normalization to 100 wt%, and Centered Log-Ratio (CLR) transformation for distance-based models. Model selection involved six classical ML algorithms (PCA-LDA, DT, RF, XGB, KNN, SVM) with rigorous hyperparameter tuning via stratified ten-fold cross-validation and grid/random search. Performance was assessed through accuracy, precision, recall, F1-scores, and external validation on an independent dataset (59 samples) to ensure generalizability.
95.8%
Classification Accuracy Achieved by Tree-Based Models (RF, XGB)
Enterprise Process Flow
Data Compilation & Filtering (XRF)
→
Data Normalization (CLR)
→
Model Selection & Hyperparameter Tuning
→
Internal Validation (Train/Test)
→
External Validation
→
SHAP Explainability Analysis
| Feature | Explainable ML Approach | Traditional Typological Approach |
|---|---|---|
| Classification Basis |
|
|
| Objectivity |
|
|
| Interpretability |
|
|
| Handling Variability |
|
|
Insights on Korean Ceramic Classification
White porcelain was accurately identified across all models due to its distinct chemical composition, characterized by minimal amounts of coloring oxides. This purity reflects specific production techniques and raw material selection.
In contrast, Celadon and Buncheong showed partial misclassification. This is attributed to their overlapping chemical characteristics, including similarities in raw materials and firing conditions (e.g., temperatures exceeding 1200°C), and historical production continuity. This highlights the nuanced relationship between these ceramic types.
Fe2O3 and TiO2 were consistently identified as the most influential components for type differentiation through SHAP analysis, aligning with established ceramic coloration mechanisms. These oxides play a crucial role in the greenish hue of celadon and the brownish tones of buncheong, making them key discriminators.
Calculate Your Potential AI ROI
Estimate the efficiency gains and cost savings your enterprise could realize by implementing similar AI-driven classification and analysis systems.
Estimated Annual Savings
Hours Reclaimed Annually
Your AI Implementation Roadmap
A typical phased approach to integrate advanced AI solutions into your enterprise workflow, ensuring a smooth transition and measurable impact.
Phase 1: Discovery & Strategy
Conduct a deep dive into existing data, infrastructure, and business goals. Define clear objectives and a tailored AI strategy based on your unique needs and the insights from this analysis.
Phase 2: Data Engineering & Model Development
Prepare and clean your proprietary datasets. Develop and train custom AI models, leveraging techniques like explainable ML to ensure transparency and trust in predictions.
Phase 3: Integration & Deployment
Seamlessly integrate the AI solution into your existing systems. Deploy models into a production environment, ensuring scalability, security, and real-time performance.
Phase 4: Monitoring & Optimization
Continuously monitor model performance, gather feedback, and iterate on improvements. Ensure the AI system evolves with your business, delivering sustained value and impact.
Ready to Transform Your Enterprise with AI?
Connect with our AI specialists to explore how these advanced methodologies can be tailored to your specific challenges and opportunities. Book a free consultation today.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat