Machine Learning
Explainable AI & Fairness Analysis
Machine-learning development and evaluation focused on predictive performance, demographic fairness, explainability, and responsible communication of results.
01
The problem
A model can achieve strong overall performance while producing uneven outcomes across groups. High-stakes use requires both predictive evaluation and a clear understanding of who may be affected differently.
02
My role
- Prepared and analyzed a large-scale justice-related dataset
- Built and compared machine-learning models
- Evaluated accuracy, recall, ROC-AUC, and group-level outcome differences
- Used SHAP and explainable-AI methods to interpret model behavior
- Presented findings for technical and non-technical audiences
03
The approach
The work paired model-performance testing with fairness analysis rather than treating accuracy as the only measure of quality. Explainability was used to inspect feature influence and communicate limitations responsibly.
04
What it demonstrates
The project strengthened practical experience in Python, data preparation, model comparison, evaluation metrics, explainability, fairness analysis, and responsible AI communication.
05
What's next
Improve reproducibility, expand model documentation, test mitigation strategies, and package the analysis into a concise public case study without exposing sensitive data.