Pottery Classification Model

Model Description

Task: Multiclass Ceramic Pottery Classification
Architecture: Multilayer Perceptron (MLP)
Input Features: 95 archaeological ceramic features
Output Classes: 9 distinct pottery types

Model Details

Performance: 99.31% Test Accuracy
Test Loss: 0.0376
Training Epochs: 500
Batch Size: 256

Dataset

Name: Ceramics: Temporal-Spatial Dataset Year: 1988 Source: Digital Archaeological Record (tDAR) Identifier:

tDAR ID: 6039
DOI: 10.6067/XCV8TD9WNB

Description: Archaeological ceramic dataset containing 95 features across 9 distinct pottery types, collected to analyze spatial and temporal characteristics of ceramic artifacts.

Features:

Total features: 95 after encoding
Feature selection process: Detailed in companion EDA Notebook

Feature	Description
firing	Firing atmosphere
temper	Type of temper used
manipul	Surface manipulation
compact	Surface compaction
color	Paint colors used
pnttype	Paint type (organic, mineral, clay)
cover	Surface slips/coatings
ware	Manufacturing technique groups
form	Vessel shape/type
culcat	Cultural Category

Classes: 9 pottery types

Training Methodology

Optimizer: Adam
Learning Rate: Initial 0.01 with exponential decay
Regularization:
- He Weight Initialization
- L2 Regularization
- Early Stopping

Intended Use

Archaeological ceramic type classification
Research in archaeological artifact analysis
Pottery provenance studies

Limitations

Trained on a specific archaeological dataset
Performance may vary with different ceramic collections
Requires careful preprocessing of input features

Citation

If you use this model, please cite: samanthajmichael/2025

License

MIT