🤖 Model Performance
Algorithm
XGBa · XGBb · LGB · CAT → Ridge
XGB · LGB · CAT → Ridge
Features
134 structural + QoL + market + temporal
149 spatial + macro + POIs + temporal lags
Base learners
4 models (XGBa, XGBb, LGB, CatBoost)
4 models (XGBa, XGBb, LGB, CatBoost)
Meta-learner
Ridge regression
Ridge regression
Cross-validation
Spatial GroupKFold · 10 folds · by NTA
Spatial GroupKFold · 5 folds · by district
OOF R² (in-sample)
—
0.9348 OOF CV
📊 Dataset
Training rows
157,329 sales records
5,531 Suhail transactions
Holdout rows
27,763 (newest 15% by date)
1,730 (2025 Q1–Q3)
Years covered
2022 – 2026
2025 – 2026
Data frequency
Individual property sales
Quarterly district aggregates
Currency / unit
USD · total property price ($)
SAR · per square metre (﷼/m²)
Median price
$820,000
﷼3,303 / m²
Typical range (P10–P90)
$304K – $2.6M
﷼1,040 – ﷼7,269 / m²
🗺 Geography
Administrative units
5 boroughs · 195 NTAs
147 districts
Primary data source
NYC Open Data · PLUTO · ACRIS
Saudi Open Data Portal · REGA
Spatial enrichment
OSM POIs · NTA polygons · NYPD
OSM POIs · Riyadh metro/bus GTFS
Map coverage bbox
40.48°N – 40.92°N · 74.26°W – 73.70°W
24.35°N – 25.10°N · 46.30°E – 47.20°E
🏗 Feature Engineering
Structural
✓ sq ft, age, floors, units, bldg class
✓ area (m²), property type
Transit
✓ subway distance + density
✓ metro dist/density · bus stops
Air quality
✓ PM2.5, NO₂ (NTA mean)
✓ PM10, NO₂ (nearest sensor)
QoL / safety
✓ crime, noise, HPD, 311 livability
— not available in open data
POIs (amenities)
✓ restaurants, bars, grocery, gym, Airbnb
✓ mosques, malls, schools, hospitals, parks
Green space
✓ tree count (200m), park proximity
✓ park POI count
Market / price history
✓ prior sale price · appreciation rate
✓ district price trend · REI index
Macroeconomic
— not included
✓ CPI, GDP, oil price, interest rates
Road connectivity
—
✓ intersection density · connectivity score
🏠 Property Types
Supported types
9 classes: Single Family, 2-Family, 3-Family, Row House, Condo, Elevator Apt, Walk-up, Co-op, Mixed Use
4 types: شقة (Apt) · فيلا (Villa) · قطعة أرض (Plot) · عمارة (Bldg)
Total bldg codes
200+ NYC building class codes
4 standardised Arabic categories
🗺 Interactive Map Layers
Choropleth layers
20 NTA layers (income, crime, noise, air…)
10 district layers (metro, price, air…)
Market overlay
— (sales in result panel)
🏠 Live listings cluster layer
Out-of-bounds mask
✓ red highlight outside NYC bbox
✓ red highlight outside Riyadh bbox