Comparing Machine Learning Approaches for Ethiopian Real Estate Price Prediction

This paper compares four machine learning models for predicting residential property prices in Addis Ababa, Ethiopia. The models tested are Linear Regression, Ridge Regression, Random Forest, and Gradient Boosting, evaluated using MAE, RMSE, R2, and 5-fold cross-validation. The study uses a dataset of 500 residential properties across 19 Addis Ababa neighborhoods, built from 2025-2026 market reports. Each property has 14 features covering size, condition, age, location, and proximity to key infrastructure. Gradient Boosting performed best with an R2 of 0.962 and a mean absolute error of 2.67 million ETB, reducing prediction error by around 59% compared to linear models. The results show that neighborhood and house area are the strongest predictors of price in the Addis Ababa market. Factors like diaspora investment, ring road access, and limited new housing supply give this market characteristics that are different from the Western markets that most existing research focuses on. The dataset, trained models, and a live Streamlit web app are all publicly available on GitHub for anyone who wants to reproduce the results or build on this work.