README.md

---
license: cc-by-4.0
task_categories:
- tabular-classification
language:
- en
tags:
- medical
- breast-cancer
- african-physiognomy
- healthcare-bias
- representation-bias
- oncology
size_categories:
- n<1K
---

# Breast Cancer Wisconsin Dataset: African Physiognomy Adjusted

## Dataset Description

This dataset addresses representation bias in medical AI by providing an African physiognomy-adjusted version of the classic Wisconsin Breast Cancer Dataset. The adjustment methodology systematically modifies cellular morphology features to better reflect documented physiological differences in African populations.

### Dataset Summary

- **Original Dataset**: Wisconsin Breast Cancer Dataset (569 samples)
- **Adjusted Dataset**: African physiognomy-adjusted version (569 samples)
- **Features**: 30 morphological features from Fine Needle Aspirate (FNA) samples
- **Classes**: Malignant (M) and Benign (B)
- **Adjustment Focus**: 212 malignant cases modified based on African physiognomy patterns

### Key Adjustments Applied

1. **Higher Breast Density**: 5-8% increase in radius, area, and texture features
2. **Enhanced Irregularity**: 12-19% increase in concavity, concave points, and fractal dimension
3. **Reduced Boundary Smoothness**: 10-12% decrease in smoothness and symmetry features

## Dataset Structure

```
breast_cancer_original.csv          # Original Wisconsin dataset
breast_cancer_african_adjusted.csv  # African physiognomy-adjusted dataset
```

### Features

Each dataset contains 32 columns:
- `id`: Sample identifier
- `diagnosis`: M (Malignant) or B (Benign)
- 30 morphological features organized as:
  - **Size/Density**: radius, area, perimeter (mean, se, worst)
  - **Texture**: texture variability (mean, se, worst)  
  - **Irregularity**: concavity, concave points, fractal dimension (mean, se, worst)
  - **Boundary**: smoothness, symmetry (mean, se, worst)
  - **Compactness**: compactness (mean, se, worst)

## Usage

### Loading the Dataset

```python
from datasets import load_dataset

# Load original Wisconsin version
original_dataset = load_dataset("electricsheepafrica/breast-cancer-african-adjusted", "wisconsin_breast_cancer_dataset")

# Access the data
original_dataset = original_dataset['train']  # Original Wisconsin dataset
```

### Example Usage

```python
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier

# Load the adjusted dataset
df = pd.read_csv("breast_cancer_african_adjusted.csv")

# Prepare features and target
X = df.drop(['id', 'diagnosis'], axis=1)
y = df['diagnosis']

# Train-test split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Train model
model = RandomForestClassifier(random_state=42)
model.fit(X_train, y_train)

# Evaluate
accuracy = model.score(X_test, y_test)
print(f"Accuracy: {accuracy:.3f}")
```

## Methodology

### Physiognomy-Based Adjustments

The adjustment methodology is based on documented physiological differences in African populations:

- **Higher Breast Density**: African women demonstrate significantly higher mammographic breast density
- **Younger Onset**: Breast cancer incidence peaks in the 40s vs 60s in Caucasian populations
- **Aggressive Subtypes**: 20-40% triple-negative breast cancer rate vs 10-15% in Caucasian populations
- **Genetic Variants**: Higher prevalence of BRCA1/2 mutations affecting cellular morphology

### Statistical Approach

- Adjustments applied exclusively to malignant cases (n=212)
- Random multipliers drawn from normal distributions
- Literature-informed scaling factors with biological variation
- Maintained feature correlations and statistical validity

## Validation and Limitations

### Validation Framework

- **Internal Validation**: Statistical consistency and biological plausibility maintained
- **External Validation Required**: Clinical validation through prospective studies in African healthcare settings

### Limitations

- Synthetic adjustments based on literature review, not direct measurement
- Limited to morphological features (excludes genetic/molecular markers)
- Single adjustment model may not capture full population diversity
- Requires validation against real African datasets

## Ethical Considerations

- Risk of perpetuating stereotypes if not properly validated
- Importance of community engagement in African healthcare settings
- Need for local data ownership and governance
- Interim solution while working toward comprehensive data collection

## Citation

If you use this dataset, please cite:

```bibtex
@misc{udodi2025breast,
  title={Addressing Representation Bias in Breast Cancer Datasets: A Physiognomy-Informed Approach for African Populations},
  author={Kossiso Udodi Royce},
  year={2025},
  publisher={Electric Sheep Africa},
  url={https://huggingface.co/datasets/ElectricSheepAfrica/breast-cancer-african-adjusted}
}
```

## Original Dataset Citation

```bibtex
@misc{street1993nuclear,
  title={Nuclear feature extraction for breast tumor diagnosis},
  author={Street, W Nick and Wolberg, William H and Mangasarian, Olvi L},
  booktitle={IS\&T/SPIE International Symposium on Electronic Imaging: Science and Technology},
  volume={1905},
  pages={861--870},
  year={1993}
}
```

## License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

## Contact

For questions or collaborations:
- **Author**: Kossiso Udodi Royce | kossi@electricsheep.africa
- **Organization**: Electric Sheep Africa
- **Year**: 2025

## Acknowledgments

- Original Wisconsin Breast Cancer Dataset creators

---

**Disclaimer**: This synthetic dataset is intended for research purposes to highlight representation bias in medical AI. Clinical validation through prospective studies in African healthcare settings is essential before any clinical application.