Project Overview
Built a classification pipeline to score business risk, with strong preprocessing controls and explainability for stakeholder reviews.
ML Model Training for Business Risk Scoring
End-to-end machine learning workflow from raw data cleaning to validated risk prediction model selection.
Key Metrics
0.87 ROC-AUC
5-fold CV
20% fewer false positives
Workflow
- Handled missing values, outliers, and standardization.
- Engineered behavioral features for model learning quality.
- Benchmarked multiple classifiers and tuned Random Forest.
Outcomes
- Improved model reliability with robust preprocessing.
- Created reusable training benchmark workflow.
- Documented feature importance for business interpretation.
Tech Stack
Python
Pandas
NumPy
Scikit-learn
Random Forest
EDA