An Implementation of the Biswas et al. (ICSE '22) Pipeline Taxonomy
This project identifies optimal retail locations in Philadelphia by implementing a rigorous 11-stage data science pipeline. The engine evaluates urban sites using a normalized suitability equation that weights transit accessibility against market competition. 🏗 The 11-Stage Pipeline Architecture
Following the study by Biswas, Wardat, and Rajan, the engine is decoupled into 11 distinct stages to ensure academic and functional reproducibility:
[ACQ] Data Acquisition: Hydration of SEPTA Transit GeoJSON and Google Places API grid-searches.
[PRP] Data Preparation: Coordinate standardization to EPSG:2272 and coordinate flattening for Web GIS.
[STR] Data Storage: Persistence of spatial primitives into a SQLite-backed GeoPackage (.gpkg).
[FTA] Feature Analysis: Weighting rail nodes vs. bus nodes and calculating competition intensity.
[TRS] Feature Transformation: Min-Max scaling and vector normalization.
[SLT] Feature Selection: Filtering assets based on data quality (e.g., minimum review thresholds).
[MDL] Modeling: Solving the Suitability Equation: S=0.7T−0.3C.
[EVL] Evaluation: Statistical validation and scheduled sensitivity analysis (April 11).
[INT] Interpretation: Quantile-binning scores into categorical Market Grades (A–D).
[CMN] Communication: Automated state reporting and terminal-based summaries.
[DPL] Deployment: A live Node.js portal rendering results via Leaflet.js.
The engine operates on a normalized model optimized for high-density business districts (e.g., Center City Philadelphia), where residential population is treated as a uniform constant.
S=0.7T−0.3C*
Transit Score (T): Quantifies accessibility via high-frequency rail and bus nodes.
Competition Intensity (C): Quantifies market resistance based on competitor density and volume.
Engine: Python (GeoPandas, NumPy, Google Maps API)
Persistence: SQLite / GeoPackage (OGC Standard)
Web Portal: Node.js, Express, EJS
Mapping: Leaflet.js
- Run Pipeline
python main.py- Launch Deployment (DPL)
cd web-portal
npm install
node server.js Biswas, S., Wardat, M., & Rajan, H. (2022). The art and practice of data science pipelines: A comprehensive study of data science pipelines in theory, in-the-small, and in-the-large. ICSE '22.