Trends, Challenges, and Future Directions in Machine Learning-Based Housing Price Prediction

Abstract

Housing price prediction is a field that is of growing importance in research, given the economic implications of the field as well as the intrinsic complexity of the real estate markets. Conventional statistical and econometric approaches, e.g., linear regression and time series analysis, usually find it hard to reflect the nonlinear, multifactorial relationship that affects property values. With the rise of machine learning (ML) methods, predictive modeling has changed as now flexible methods that are data-oriented and can work with large and diverse datasets are possible. Decision trees and random forests, gradient boosting (XGBoost), and neural networks have proven to be more accurate predictors when compared to traditional models. The latest trends focus on ensemble and hybrid modeling, the incorporation of alternative sources of data, such as satellite imagery and socioeconomic factors, and the implementation of explainable AI (XAI) to promote transparency. Regardless of these developments, various challenges do exist, such as data quality, interpretation of the model, and dynamism of the housing markets. This paper presents an extensive literature analysis of the field of ML-based housing price prediction, including the attributes of properties that cause the greatest impact on their price, the methodological strategies that are pertinent, the current trends, and the principal limitations that limit the predictive accuracy, to provide an idea of the future studies and possible applications.