London Real Estate Development | Predictions & Recommendations

Sherlock Homes is a data-driven Real Estate project focused on the London market. By combining digitalization with construction insights, we developed Watson V1.0, a machine learning model that predicts dwelling prices and acts as a strategic advisor to identify high-ROI renovation opportunities.

• Web App: Sherlock Homes App
• Tableau Dashboard: London Real Estate Analysis
• Project Presentation: Canva Slides

• Market Value: The London residential market is valued at approximately £2.64 trillion.
• Stability: After a correction in 2023 due to mortgage rate surges, the market has stabilized.
• Current State: As of late 2025 data, the average property price is £652,000, with median prices around £505,000.

The dataset aggregates two primary sources (filtered for data > 2023):

1. London House Price Data (Kaggle): Transactional data, rooms, tenure, energy rates.
2. London Building Stock Model (London Datastore): Conservation area status, building age bands.

Processing:

• Cleaned ~4.3M raw rows down to ~80,000 high-quality rows for training.
• Feature Engineering: Added neighbourhood median values (£/m²) and size buckets.
• Handling Missing Data: Null room counts replaced by the median of similar properties; outliers (e.g., short leaseholds <18 years) removed.

We experimented with several algorithms to optimize prediction accuracy. LightGBM was selected as the final model due to its superior performance and speed.

Model Specs:

• Target: Logarithmic transformation of Selling Price.
• Encoders: One-Hot (Property Type, Tenure), Label (Energy), Target Encoding (Location).

The Streamlit app (watsonV1_0.py) allows users to input property details and receive:

1. Valuation: Estimated market value with a confidence interval (25%-75% variance).
2. Sherlock's Strategic Advisor:
   • Feasibility Check: Ensures recommended layouts (e.g., adding a bedroom) fit within the floor area.
   • Market Reality Check: Verifies that the proposed layout actually exists among the nearest 50 neighbors.
   • ROI Logic: Only recommends changes if the predicted value uplift is >10%.

• 1_Data/: Contains the processed dataset (df_cleaned.csv).
• 2_Notebooks/:
  • Sherlock_Homes-Analysis.ipynb: Exploratory Data Analysis (EDA).
  • LBSM_datacleaning.ipynb: Cleaning external building stock data.
  • Sherlock_Homes - ML-RF...: Random Forest training.
  • Sherlock_Homes - ML-XG...: XGBoost training.
  • Sherlock_Homes - ML-LGBM: Final LightGBM model training.
  • Watson_LGBM.pkl: The pickled model used by the app.
• 3_App/:
  • watsonV1_0.py: The main application script.
  • requirements.txt: List of dependencies.
  • SH2.png: Assets/Images.

To run the application locally:

1. Clone the repository:

   git clone https://github.com/YourUsername/SherlockHomes.git
   cd SherlockHomes

2. Install dependencies: (Note: The requirements file is located inside the 3_App folder)

   pip install -r 3_App/requirements.txt

3. Run the App:

   streamlit run 3_App/watsonV1_0.py

António Galvão