Smart House Price Predictions using YDF

Abstract

Predicting housing prices is a complex task within data science and real estate analytics due to intricate relationships among economic, structural, and locational factors. These relationships usually cannot be captured by traditional statistical regression methods due to their nonlinear nature. This paper proposes a robust and interpretable regression framework using Yggdrasil Decision Forest-one of the advanced ensemble learning systems developed by Google. In this work, a Random Forest regressor is implemented in YDF, which can handle heterogeneous data with minimal preprocessing due to its intrinsic nature. The performance of the model was gauged through RMSE, MAE, and the Coefficient of Determination, R². To provide a benchmark for the performance, the proposed model was compared against Linear Regression, Decision Tree, and Gradient Boosted Tree (YDF) models. Experimental results showed that the random forest coupled with YDF yielded the best predictive performance, with a coefficient of determination of 0.9947 and lowest RMSE and MAE compared with all other baseline models. This therefore validates the efficiency, scalability, and strength of generalization of the YDF framework, hence solidifying its potential to become a reliable and high-performance solution for real-world residential price prediction tasks.