Machine Learning Report

Dual Model Architecture for Real Estate Predictions & Recommendations

MAPE ±20% Dual Models XGBoost 3.0.0
Project Overview
  • Problem Type: Regression
  • ML Approach: Supervised
  • Business Objective: ±20% MAPE
System Architecture

Data → Preprocessing → Model → Prediction (API)

Model Details
  • Version 3.0.0
  • Predictor Model XGBoost
  • Training Date: April 5,2025

About Dataset

Data Sources
  • Primary: Kaggle
  • Dataset: Link
  • 168K+ Property Listings
Data Statistics
  • Total Samples 168,446
  • Features 20
  • Data Validity 2018-2019

Data Analysis & EDA

Key Findings
  • Bimodal Price Distribution
  • 72% For Sale Properties & 28% Rental Properties
  • Strong Correlation of Purpose with Target Feature
Data Cleaning
Outliers Removed 3.78%
Unvalid Data Removed 24.5%
Duplicate Data Removed 42%

Data Preparation & Preprocessing

Raw Data → Duplicated Data Removal → Outlier Detection + Removal Using Z-Method→ Unvalid Data Removal → Cleaned Data for Model Building
Key Transformations
  • Feature Engineering
  • Ordinal Encoding + Target Encoding for Categorical Features
  • Standard Scaling for Numerical Features

Model Selection

Sale Prediction Model
Model MAE R²
XGBoost 4.5M 0.86
Random Forest 4.6M 0.85
Gradient Boosting 4.7M 0.84
Linear Regression 5.3M 0.78
Rent Prediction Model
Model MAE R²
XGBoost 22.0K 0.82
Random Forest 24.7K 0.81
Gradient Boosting 27.5K 0.80
Linear Regression 38.0K 0.72

Training Pipeline

Training Pipeline

Dual Prediction Models

Sale Prediction
XGBoost Regressor MAPE 23%
  • n_estimators: 126
  • max_depth: 11
  • learning_rate: 0.123
Rent Prediction
XGBoost Regressor MAPE 21%
  • n_estimators: 137
  • max_depth: 10
  • learning_rate: 0.047

Model Evaluation

Sale Prediction Model
MAE
4.5M
PKR
MAPE
23%
Error
R²
0.86
Score
Success: Acheived Approx MAPE of ±20%
Rent Prediction Model
MAE
22.0K
PKR
MAPE
21%
Error
R²
0.82
Score
Success: Acheived Approx MAPE of ±20%
Sale Recommender Model
AVG Similarity
92.7%
Coverage
13.1%
Novelty
52.6%
Success Rate: 100%
Rent Recommender Model
Avg Similarity
81.9%
Coverage
32.6%
Novelty
46.9%
Success Rate: 100%

Hyperparameter Tuning Using Optuna

Sale Prediction Model
  • Optimization Method Bayesian
  • Iterations 100
  • Best Parameters n_estimators: 126 max_depth: 11 learning_rate: 0.123
  • Best Accuracy 87%
Rent Prediction Model
  • Optimization Method Bayesian
  • Iterations 100
  • Best Parameters n_estimators: 137 max_depth: 10 learning_rate: 0.047
  • Best Accuracy 83%

Deployment

API Endpoints 
Production-ready ML models deployed as RESTful API endpoints with JSON payload support.

Example Request

POST < api-url >/predict
Content-Type: application/json

{
    "purpose": "rent",
    "property_type": "House",
    "location": "DHA Defence Islamabad",
    "city": "Islamabad",
    "province_name": "Islamabad Capital",
    "baths": [3],
    "bedrooms": [4],
    "area": [8],
    "Area Category": "5-10 Marla"
}

Success Response


{
    "predicted_price": 47243,
    "recommendations": [
        {
            "Area Category": "1-5 Kanal",
            "area": 80.0,
            "baths": 3,
            "bedrooms": 3,
            "city": "Islamabad",
            "location": "F-7",
            "price": 200000,
            "property_type": "House",
            "province_name": "Islamabad 
        },
        {
            "Area Category": "1-5 Kanal",
            "area": 72.0,
            "baths": 4,
            "bedrooms": 4,
            "city": "Islamabad",
            "location": "F-8",
            "price": 170000,
            "property_type": "House",
            "province_name": "Islamabad Capital"
        },
        {
            "Area Category": "1-5 Kanal",
            "area": 72.0,
            "baths": 4,
            "bedrooms": 4,
            "city": "Islamabad",
            "location": "E-11",
            "price": 50000,
            "property_type": "Upper Portion",
            "province_name": "Islamabad Capital"
        }
    ]
}
Visit GitHub