EXP-007 / question
Technical question
Can SQL features and interpretable regression support site-selection decisions?
EXP-007 / method
Method and workflow
- Generate or load location-level business data.
- Create modeling features through SQL transformations.
- Train and compare regression models against simple baselines.
- Evaluate holdout metrics and cross-validation behavior.
- Score candidate shop locations and explain key profit drivers.
- Save outputs and diagnostics for review.
location data
SQLite features
model comparison
validation
diagnostics
candidate scoring
EXP-007 / evidence
Evidence of work
Business framing
Predictions are translated into candidate-location ranking rather than only reporting a model score.
Model honesty
Baseline checks, CV, and residual diagnostics make performance easier to interpret.
Engineering
SQLite features and repeatable scripts make the workflow more reproducible than a notebook-only demo.
EXP-007 / stack
Technical stack
PythonpandasNumPySQLitescikit-learnmatplotlibjoblibunittestRuffBlackmypyGitHub Actions
Open repository ↗
EXP-007 / limitations
Limitations and honesty check
- The data is synthetic or demo-oriented, so results should not be treated as real market advice.
- Site selection would require lease terms, foot traffic, competition, local events, and operational constraints.
- The model supports decision review; it should not replace business judgment.
EXP-007 / next
Next improvements
- Add richer geospatial features and local competition signals.
- Add scenario analysis for rent, labor, and demand shifts.
- Add a small dashboard for candidate comparison.
- Add more robust model cards and data documentation.