maryzhang/hw1-24679-tabular-dataset

Shoe Size Measurements Tabular Dataset

Dataset Summary

Purpose: This dataset was created for tabular data analysis and prediction tasks involving shoe measurements, developed as part of CMU 24-679 coursework to explore tabular data augmentation techniques.

Quick Stats:

• 338 total samples (30 original + 308 augmented)
• 3 numerical features + 3 categorical features
• High correlation between size measurements (>0.97)
• ~10x augmentation factor

Contact: maryzhang@cmu.edu

Exploratory Data Analysis

Summary Statistics (Original Data)

Statistic	US Size	Shoe Size (mm)	Actual Length (mm)
count	30.0	30.0	30.0
mean	5.8	227.7	230.6
std	0.7	5.8	6.2
min	5.0	218.0	222.0
25%	5.0	223.0	225.0
50%	6.0	230.0	230.0
75%	6.0	230.0	235.0
max	7.5	240.0	240.0

Visualizations

Key Findings

• Strong positive correlation (>0.85) between all size measurements
• Distribution centered around US size 6 (women's sizing)
• No missing values in dataset
• Size range: US 5.0 to 7.5 (women's shoe sizes)
• Most common brands: Nike (5), Steve Madden (4), Adidas (4)
• Shoe types: Sneakers (10), Slippers (9), Heels (6), Loafers (3), Boots (2)
• Colors: Black (9), White (6), Pink (2), Grey (2), others (11)

Dataset Composition

Features

• US size: Integer (US shoe size, 6-13)
• Shoe size (mm): Integer (manufacturer size in mm)
• Actual measured shoe length: Integer (measured length in mm)
• Type of shoe: String (Sneakers, Boots, Dress Shoes, Athletic)
• Shoe color: String (Black, White, Brown, Gray, Other)
• Shoe Brand: String (Nike, Adidas, Vans, Converse, etc.)

Distribution

Category	Values	Most Common
Type	4 unique	Sneakers (50%)
Color	5 unique	Black (40%)
Brand	6 unique	Nike (27%)

Data Splits

• original: 30 manually collected measurements
• augmented: 308 synthetically generated samples

Data Collection Process

Collection Methodology

Data collected January-February 2025:

• Personal shoe collection measurements
• Retail store size charts
• Manufacturer specifications
• Manual measurements using standard ruler

Selection Criteria

• Common shoe brands and types
• Adult sizes (US 6-13)
• Verified measurements
• Diverse shoe categories

Preprocessing and Augmentation

Preprocessing Pipeline

1. Converted all measurements to integers (3-digit mm)
2. Standardized categorical values
3. Removed any invalid entries
4. Verified measurement consistency

Augmentation Techniques

Generated ~10x samples using:

• Gaussian Noise: ±5% variation on numerical features
• Linear Interpolation: Between similar sizes
• SMOTE-inspired: Synthetic samples between neighbors
• Category Shuffling: 10% probability of categorical variation
• Correlation Preservation: Maintained size relationships

Labels and Annotation

Measurement Schema

• All numerical features in millimeters
• US sizes follow standard conversion
• Actual length may vary ±5mm from manufacturer size

Data Quality

• No missing values
• All correlations preserved in augmentation
• Categorical distributions maintained

Intended Use and Limitations

Intended Use Cases

• Shoe size prediction models
• Tabular augmentation studies
• Regression/classification tasks
• Educational demonstrations
• Size conversion algorithms

Limitations

• Small original dataset (30 samples)
• Limited to adult sizes
• May not represent all brands equally
• Augmented data maintains original distributions
• Not suitable for medical/orthopedic applications

Out-of-Scope Uses

• Medical foot assessments
• Children's shoe sizing
• Production quality control
• Commercial sizing systems without validation

Ethical Considerations

Representation

• Limited to common US sizes
• May not represent global sizing standards
• Brand selection based on availability

Privacy

• No personal information included
• Measurements anonymized
• No individual foot measurements

Bias Considerations

• Western brand bias
• Adult size focus
• May not represent specialty footwear

AI Usage Disclosure

AI-Assisted Components

• Augmentation code: Partially generated with AI assistance
• Statistical analysis: Manual calculations
• Documentation: Structure refined with AI help
• Data: All measurements manually collected

Human Oversight

• All original data manually measured
• Augmentation parameters tuned empirically
• Quality checks performed on all samples
• Final dataset manually reviewed

Usage Example

from datasets import load_dataset
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor

# Load dataset
dataset = load_dataset("maryzhang/hw1-24679-tabular-dataset")

# Convert to DataFrame
df = pd.DataFrame(dataset['augmented'])

# Prepare features
X = df[['US size', 'Shoe size (mm)']]
y = df['Actual measured shoe length']

# Train model
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
model = RandomForestRegressor()
model.fit(X_train, y_train)

# Evaluate
score = model.score(X_test, y_test)
print(f"R² Score: {score:.3f}")
## Exploratory Data Analysis

Citation

bibtex@dataset{zhang2025shoe, author = {Mary Zhang}, title = {Shoe Size Measurements Tabular Dataset}, year = {2025}, publisher = {Hugging Face}, note = {CMU 24-679 Homework 1}, url = {https://huggingface.co/datasets/maryzhang/hw1-24679-tabular-dataset} }

License

This dataset is released under the MIT License.

Contact

Dataset created by Mary Zhang for CMU 24-679. For questions or issues, please contact maryzhang@cmu.edu.