HamzaEric/Kenyan-Real-Estate-Market-Valuation

An end-to-end machine learning application that predicts residential property prices in Kenya. It uses a tuned LassoCV model with cross-validation to improve accuracy and reduce overfitting. # Houses Price Prediction ## Kenyan Real Estate Valuation Engine An end-to-end machine learning application that predicts residential property market values in Kenya using regularized linear regression modeling. The architecture features an automated data preprocessing pipeline and a cascading user interface built with Streamlit, deployed to serve real-time baseline valuations. --- ## Project Architecture & Machine Learning Pipeline This repository transitions a raw real estate dataset through feature engineering, rigorous statistical optimization, and web serialization: 1. **Data Cleaning & Harmonization:** Standardizes noisy, street-level geographic fields using regex-based cleaning operations. 2. **Feature Engineering:** Maps precise local coordinates and neighborhoods into broad macroeconomic development sectors (**Zoning Tiers**) using deterministic rule-based mapping blocks. 3. **Preprocessing Pipeline:** Employs a Scikit-Learn `ColumnTransformer` to seamlessly apply `OneHotEncoder` to categorical string metrics and scale numerical dimensions. 4. **Regularization Optimization:** Combines 5-fold cross-validation with an $L_1$ penalty (**LassoCV**) to drive redundant, low-signal sparse feature weights completely to absolute zero, successfully defeating the Bias-Variance tradeoff. ### Model Evaluation Scoreboard Our structural validation curves yielded an incredibly tight generalization profile across linear strategies: | Model Strategy | Train RMSE | Test RMSE | Generalization Gap (%) | Status | | :--- | :--- | :--- | :--- | :--- | | **Simple Mean Baseline** | 40,183,897.19 KSh | 40,183,897.19 KSh | 0.00% | Benchmark Floor | | **Standard Linear Regression** | 24,993,041.79 KSh | 25,084,263.00 KSh | 0.3650% | Unconstrained Baseline | | **Optimized RidgeCV ($\alpha=10$)** | 25,106,404.76 KSh | 25,294,474.21 KSh | 0.7491% | Over-smoothed Weights | | **Optimized LassoCV ($\alpha=1000$)** | **24,993,120.13 KSh** | **25,078,562.37 KSh** | **0.3419%** | **Pr …