🏠 Egypt Real Estate Price Predictor

A machine learning model trained on 18,838 Egyptian property listings to predict real estate prices across Egypt.

Model Details

Property	Value
Algorithm	XGBoost (gradient boosting)
Target	log1p(price in EGP)
R² Score (log scale)	0.647
Median APE	27.5%
Training samples	18,838
Cities covered	16 Egyptian governorates

Features Used

Feature	Description
`size_sqm`	Property area in square meters
`bedrooms_clean`	Number of bedrooms
`bathrooms`	Number of bathrooms
`city`	Governorate (Cairo, Giza, North Coast, etc.)
`property_type`	Apartment, Villa, Chalet, Duplex, etc.
`is_installment`	Payment method (0=Cash, 1=Installments)
`city_median_price`	Median price in that city (target encoding)
`type_median_price`	Median price for that property type
`bed_bath_ratio`	Bedrooms / Bathrooms ratio
`rooms_total`	Total rooms count
`size_per_room`	Average sqm per room

Cities Covered

Cairo · Giza · North Coast · Red Sea · Alexandria · Suez · Qalyubia · South Sinai · Matrouh · Al Daqahlya · Aswan · Asyut · Damietta · Luxor · Sharqia · Kafr El Sheikh

Property Types

Apartment · Villa · Chalet · Duplex · Townhouse · Twin House · Penthouse · Hotel Apartment · Other

Pipeline

The model uses a sklearn.pipeline.Pipeline with:

ColumnTransformer — StandardScaler for numerics, OrdinalEncoder for categoricals
XGBRegressor — 1000 estimators, lr=0.02, max_depth=8

Usage

import joblib
import numpy as np
import pandas as pd
import json

# Load model and metadata
model = joblib.load("real_estate_model.pkl")
with open("model_metadata.json") as f:
    meta = json.load(f)

def predict_price(area_sqm, bedrooms, bathrooms, city, property_type, is_installment=0):
    cm = meta["city_median"]
    tm = meta["type_median"]
    rooms = bedrooms + bathrooms
    global_med = meta["global_median"]
    
    row = pd.DataFrame([{
        "size_sqm":         area_sqm,
        "bedrooms_clean":   bedrooms,
        "bathrooms":        bathrooms,
        "is_installment":   is_installment,
        "bed_bath_ratio":   bedrooms / max(bathrooms, 1),
        "rooms_total":      rooms,
        "city_median_price": cm.get(city, global_med),
        "type_median_price": tm.get(property_type, global_med),
        "size_per_room":    area_sqm / max(rooms, 1),
        "city":             city,
        "property_type":    property_type,
    }])
    
    log_price = model.predict(row)[0]
    price = np.expm1(log_price)
    return {
        "estimated_price_egp": round(price),
        "price_low_egp":       round(price * 0.90),
        "price_high_egp":      round(price * 1.15),
        "price_per_sqm_egp":   round(price / area_sqm),
    }

# Example
result = predict_price(
    area_sqm=150,
    bedrooms=3,
    bathrooms=2,
    city="Cairo",
    property_type="Apartment"
)
print(result)
# {'estimated_price_egp': 9800000, 'price_low_egp': 8820000, ...}

Data Source

Scraped from PropertyFinder Egypt. Data cleaned, outliers removed, and features engineered for optimal model performance.

Limitations

Prices reflect market listings (not sale prices)
High variance is inherent to real estate (view, floor, finish not captured)
North Coast / Red Sea prices are seasonal/resort properties

Downloads last month: -; Downloads are not tracked for this model. How to track