A spatially explicit, multi-layer machine learning framework for neighbourhood growth prediction and livability analysis across Delhi NCR — built entirely on real geospatial data.
Overview tab — trajectory map with cell-level analysis panel
IUD-NEPS scores every 500m × 500m cell across 2,913 km² of Delhi NCR on two dimensions:
- Growth Potential — where property values and economic activity are likely to increase, predicted by an XGBoost + LightGBM ensemble (R² = 0.85)
- Livability — where it's actually good to live, based on air quality, healthcare, parks, safety, and civic amenities
It then classifies each cell into one of four trajectories — Accelerating, Emerging, Stable, or Declining — and simulates the impact of proposed infrastructure like new metro lines and RRTS corridors.
- Python, GeoPandas, OSMnx, Rasterio
- XGBoost, LightGBM, SHAP
- Leaflet.js, Chart.js
- Spatial indexing + GeoJSON pipelines
- OpenStreetMap + Census + AQI datasets
Growth potential heatmap — zoom in past level 13 to see ward-level property prices. Click any cell for detailed score breakdown.
Livability score — weighted composite of AQI, healthcare, education, parks, recreation, civic amenities and crime proxy.
Multi-select infrastructure scenarios — combine multiple corridors to see cumulative impact across the city.
| Model | R² | RMSE |
|---|---|---|
| Ridge Regression | 0.77 | 0.079 |
| Random Forest | 0.81 | 0.072 |
| XGBoost | 0.85 | 0.063 |
| LightGBM | 0.85 | 0.064 |
| Ensemble (XGB + LGBM) | 0.85 | 0.063 |
Top SHAP features: working age fraction, migration rate, airport proximity, accessibility to economic centres.
| Data | Source | Type |
|---|---|---|
| Road network + POIs | OpenStreetMap (Geofabrik) | Downloaded |
| DMRC metro stations | OpenStreetMap | Downloaded |
| Property prices | DDA / public ward records | Seeded CSV |
| Census demographics | Census of India 2011 | Seeded CSV |
| Air quality (AQI) | CPCB monitoring stations | Seeded CSV |
| Crime proxy | Delhi Police / NCRB 2022 | Seeded CSV |
- Python 3.12
- ~500 MB disk space for OSM data
git clone https://github.com/yourusername/iud-neps
cd iud-neps
py -3.12 -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # Mac/Linux
py -3.12 -m pip install -r requirements.txtDownload the OSM extract for North India from Geofabrik:
https://download.geofabrik.de/asia/india/northern-zone-latest.osm.pbf
Place it in data/raw/.
Run each script in order from the project root:
# Step 2 — Extract POIs and seed CSV data
py -3.12 scripts/fetch_data.py
# Step 3 — Build grid and compute all features
py -3.12 scripts/feature_engineering.py
# Step 4 — Train XGBoost + LightGBM ensemble
py -3.12 scripts/train_models.py
# Step 5 — Classify cells into trajectories
py -3.12 scripts/classify.py
# Step 6 — Run infrastructure scenario simulations
py -3.12 scripts/simulate.py
# Step 7 — Generate interactive dashboard
py -3.12 scripts/dashboard.pyThe dashboard opens automatically in your browser and is saved to outputs/dashboard.html.
iud-neps/
├── config.py # All settings — edit here to change anything
├── requirements.txt
├── .gitignore
│
├── scripts/
│ ├── fetch_data.py # Step 2: data collection
│ ├── feature_engineering.py # Step 3: 28 spatial features
│ ├── train_models.py # Step 4: model training + SHAP
│ ├── classify.py # Step 5: trajectory classification
│ ├── simulate.py # Step 6: scenario simulation
│ └── dashboard.py # Step 7: dashboard generation
│
├── dashboard/ # Modular dashboard code
│ ├── data_prep.py
│ ├── geojson_builder.py
│ ├── html_generator.py
│ └── templates/
│ ├── css/style.css
│ └── js/
│ ├── map.js
│ ├── charts.js
│ ├── panels.js
│ └── tabs.js
│
├── assets/ # README screenshots
├── data/
│ ├── raw/ # OSM extracts + CSVs (git ignored)
│ └── processed/ # Feature matrices (git ignored)
│
├── models/
│ ├── potential_model/ # XGBoost + LightGBM + SHAP (git ignored)
│ └── livability_model/ # Weighted composite index
│
└── outputs/ # Dashboard HTML (git ignored)
- Overview — Trajectory map with cell-level analysis panel (scores, real estate, simulation impact)
- Potential — Growth heatmap with property prices at zoom 13+, SHAP feature importance
- Livability — Livability heatmap with data quality transparency per feature
- Simulations — Multi-select scenarios with combined impact overlay
- Delhi-Meerut RRTS Full Opening
- Delhi-Gurugram RRTS Corridor
- Noida Metro Phase 2 Expansion
- Delhi Ring Road Metro Corridor
All parameters live in config.py:
GRID_SIZE_M = 500 # cell size in metres
N_CV_BLOCKS = 10 # cross-validation folds
LIVABILITY_WEIGHTS = {...} # adjust livability feature weights
SCENARIOS = {...} # add or modify simulation corridors- Census data is from 2011 — migration rates may be outdated
- Property prices are ward-level approximations, not transaction-level
- Parks POI count (114) is low due to inconsistent OSM tagging in India
- Accessibility uses Euclidean distance, not actual road network travel time
- Residual spatial autocorrelation (Moran's I = 0.96) indicates missing real estate momentum features
This project is based on independent research conducted by Ashish Raymajhi at Sharda University.
Associated unpublished research manuscript: "Integrated Urban Dynamics and Neighbourhood Evolution Prediction System Using Multi-Layer Spatial-Temporal Analytics"
Ashish Raymajhi (Lead Author) — [email protected]
Dhruv Rai (Co-author) — [email protected]
This codebase extends and improves upon the original research:
- R² improved from 0.67 to 0.85
- Percentile-based trajectory classification (fixes 92%-stable problem)
- Dual scoring system (growth potential + livability)
- Airport proximity and AQI introduced as new features
- Realistic 3-zone infrastructure simulation with spillover effects
MIT License — free to use, modify, and distribute with attribution.