Comparative Performance of Selected Machine Learning Algorithms in Predicting Residential Property Price Across Sub-Markets in Lagos Metropolis

Abstract

Traditional techniques of property valuation are increasingly criticised for their subjectivity, vulnerability to data insufficiency, and other discrepancies, thereby fuelling a methodological shift towards automated and data-driven valuation frameworks. Hence, this study explores three selected machine learning algorithms - Random Forest (RF), Bayesian Ridge (BR), and LASSO Regression – in estimating residential property prices across three distinct sub-markets within Lagos Metropolis. The study seeks to identify the algorithm that achieves the highest predictive precision and to examine the extent to which sub-market heterogeneity affects model reliability. The dataset utilised for the study comprised 469 residential property transactions concluded over the years by thirteen registered Estate Surveying and Valuation firms operating within the Metropolis. The results revealed varied performance across the sub-markets, with Random Forest emerging as the most efficient model, demonstrating the model’s robustness and adaptability to nonlinear property data and spatial variability. The Bayesian Ridge Regression model produced moderate results, performing reliably in sub-markets characterised by stable property structures, while LASSO Regression yielded the least accuracy, owing to its sensitivity to linear assumptions and coefficient shrinkage. By situating algorithmic comparison within a spatially segmented urban framework, this research contributes to the development of location-sensitive, data-driven valuation models that can enhance valuation practice, improve investor confidence, and promote greater transparency and efficiency in Nigeria’s property market.