Aditya Kasturi*
Volume 1, Issue 1
Date of Publication: 31 October, 2025
In this study, we propose and evaluate a machine learning framework for predicting
high- return- on- investment (ROI) residential properties in Seattle, Washington,
drawing on both structured and unstructured data and informed by robust academic
research.
Background & Motivation
Localized forecasting models remain underdeveloped despite the rapid growth and high
volatility of Seattleās housing market. Building on prior Seattle-specific workāsuch as
Zhang (2024) which compared polynomial regression, K- nearest neighbors, and multiple
linear regression using Seattle data, finding interior living space and building design to be
significant predictors. we extend the analysis encompassing modern ensemble and
multimodal approaches.
Methods
We assemble a dataset of 4,600+ property transactions in King County from public
records (similar to the Kaggle dataset used by ResearchGate study) .Features include
size, bedrooms, lot area, ZIP code, school district rating, transit proximity, crime
statistics, and property description text. We engineer structured variables (sqft,
bedrooms, age), spatialātemporal lag features, and embed unstructured listing
descriptions using transformer-derived NLP embeddings following the multimodal deep
learning approach of Hasan et al. (2024). We train and compare several models:
Random Forest, XGBoost/Gradient Boosting, and StackingAveragedModels (the latter was
top performer in the Seattle case using R² ā 0.777, RMSLE = 0.2328). Hyperparameter
tuning uses Bayesian optimization frameworks, as recommended in Chen et al. (2023).
Model interpretability uses SHAP (Shapley additive explanations) to quantify feature
influence
Seattle Real Estate, Housing Price Prediction, Machine Learning, Ensemble Models, Multimodal Data, SHAP Interpretability, High-ROI Investment
Aditya Kasturi, Realogics Sothebyās International Realty, USA.
Aditya, K. (2025). Smart Real Estate Investment: Machine Learning Models for Identifying High-ROI Properties in Seattle. J Cogn Comput Ext Realities, 1(1), 01-15.