AI-driven forecasting strategies for real estate price prediction

Abstract

Accurately forecasting property prices is vital for decision-making in real estate markets. Traditional econometric models, valued for their transparency and regulatory credibility, are constrained by restrictive assumptions that limit their ability to capture complex dynamics. To address these shortcomings, machine learning methods have been increasingly adopted, offering greater flexibility and higher predictive accuracy. However, such models are often opaque, difficult to justify, and highly sensitive to task–model alignment and data quality. Hybrid approaches, designed to integrate complementary strengths, present a promising alternative but remain empirically scarce and unevenly evaluated. This dissertation conducts a systematic review of sixty peer-reviewed studies to assess how AI-based forecasting models are structured and classified, how they are benchmarked against traditional and machine learning baselines, and how they align with methodological requirements in residential property price prediction. Following the PRISMA protocol, the review reveals a predominance of ensemble and deep learning models, while hybrid designs are underexplored and frequently weakly justified. Benchmarking practices, though widespread, lack consistency, and interpretability tools are incorporated in only half of the studies. These findings portray a field marked by technical innovation but insufficient consolidation, where forecasting credibility depends as much on justification, transparency, and alignment as on predictive accuracy.

Date of Award	11 Dec 2025
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Aydin Teymourifar (Supervisor)