Machine Learning & Regression
Production Deployed
Ensemble XGBoost + LightGBM
Supervised Machine Learning Real Estate Valuation Engine
An enterprise-grade pricing model combining automated feature engineering, spatial clustering, Bayesian hyperparameter optimization with Optuna, and sub-20ms inference REST APIs.
R² Accuracy
0.914
Cross-validated
Inference Latency
18ms
Fast REST API
RMSE Reduction
-32%
vs Legacy Baselines
Features Engineered
35+
Spatial & Structural
The Problem
Real estate portfolios and appraisal desks struggle with outdated linear regressions that fail to model non-linear interactions between school district ratings, localized crime rates, structural depreciation, and square-footage thresholds.
The Solution
Built an ensemble gradient-boosted regression pipeline with automated target encoding, spatial distance clustering, and Bayesian Optuna hyperparameter optimization delivered via a high-performance Flask microservice.
Model Inference Class
# Supervised Gradient Boosting Inference Pipeline
import joblib
import pandas as pd
import numpy as np
class RealEstateValuator:
def __init__(self, model_artifact_path: str = "models/ensemble_pipeline.joblib"):
self.pipeline = joblib.load(model_artifact_path)
def estimate_market_price(self, property_features: dict) -> dict:
input_df = pd.DataFrame([property_features])
# Spatial and structural feature calculations
if 'sqft' in input_df and 'bedrooms' in input_df:
input_df['sqft_per_bedroom'] = input_df['sqft'] / np.maximum(input_df['bedrooms'], 1)
predicted_price = float(self.pipeline.predict(input_df)[0])
return {
"estimated_valuation": round(predicted_price, 2),
"lower_bound_90ci": round(predicted_price * 0.93, 2),
"upper_bound_90ci": round(predicted_price * 1.07, 2),
"currency": "USD"
}"Majid's predictive machine learning model helped us automate our underwriting appraisal checks with 91.4% accuracy. Fast, robust, and cleanly documented."
Jonathan L.
Portfolio Director • Canadian Real Estate Investment Trust