Giftyyyyy/predictive-analytics-project

Predictive analytics case study comparing New York and Nairobi real-estate markets for a $500M investment fund — regression modeling, SARIMA/Holt-Winters forecasting, and evidence-based investment recommendations. OMIS 308, University of Ghana Business School. # ED Happiness Estates — Predictive Analytics Investment Case Predictive analytics case study for OMIS 308 (University of Ghana Business School), evaluating where a hypothetical $500M real-estate fund should deploy capital across global cities. ## Overview ED Happiness Estates is exploring rental and sale developments in high-potential cities. This project uses cross-sectional regression and time-series forecasting to evaluate two supplied markets — **New York** and **Nairobi** — and turns the model outputs into evidence-based investment recommendations rather than speculative city rankings. ## What's in this repo - Data cleaning and feature engineering (winsorization, target-leakage controls, missing-value handling) - Regression modeling: Random Forest, Ridge Regression, Linear Regression, Decision Tree, Gradient Boosting - Time-series forecasting: SARIMA and Holt-Winters, evaluated on a 24-month holdout - A full write-up translating model results into investment guidance, with explicit limitations ## Key results | Market | Best model | Headline metric | |---|---|---| | New York (transaction price) | Random Forest | RMSE $2.35M, R² 0.10 | | New York (24-month price index forecast) | SARIMA(1,1,1)x(1,1,1,12) | MAE 1.14, RMSE 1.42, MAPE 0.35% | | Nairobi (property price) | Linear Regression | RMSE KES 2.09M, R² 0.966, CV R² 0.959 | New York's transaction model is deliberately treated as a screening benchmark, not a valuation model — its low R² reflects that key predictors (usable area, coordinates) were largely missing from the supplied data. Nairobi's model is stronger and identifies rental price, floor area, and land price as the leading valuation drivers. ## Approach A third city and a matching Kenyan macroeconomic series were not available in the supplied data (the Ghana GDP file could not be validly joined to Kenyan property records), so this project reports a two-market comparison rather than fabricating a three-city ranking. Recommendations are stage …