Hedonic methods are widely used to estimate real estate prices. However, such methods require multiple regression analysis, making it difficult to consider the complex motivations and preferences that shape real estate prices because of multicollinearity (Selim 2009; Chiarazzo et al. 2014).

Neural networks (NNs) have recently gained considerable attention as analytical methods that compensate for the shortcomings of hedonic methods. An advantage of using NNs is that they diminish the need to assume explicit functions between the input and output of studies because NNs learn directly from observed data (Selim 2009; Chiarazzo et al. 2014). As research on using NNs in real estate price estimation progresses, several studies have compared the accuracy of hedonic methods with NNs. Selim (2009) compared the accuracy of hedonic methods and N models for estimating house prices in Turkey and showed that these are a better alternative to hedonic methods.

However, using NNs also has certain disadvantages. First, the choice of explanatory variables for the model has been insufficiently examined. Previous studies often employed only micro variables as explanatory variables in their models, and few employed macro variables in their models, despite macro variables, such as GDP, significantly impacting real estate prices (Yakub et al. 2020). Population dynamics can also affect the accuracy of real estate price estimation models (Lin et al. 2018). However, few studies employed population dynamic variables in their models.

Second, similar to NNs, Random Forest (RF) has been applied to estimate real estate prices to overcome the limitations of the hedonic model. Fonseca et al. (2023) compared the accuracy of HA, NNs, and RF models for estimating house prices in Lisbon using micro variables where RF had the best accuracy. However, research on the accuracy of different combinations of explanatory variable patterns and multiple estimation algorithms remains inadequate.

Therefore, this study compares the accuracy of models using RF and NNs with micro-, macro-, and population-dynamic variables. The analysis used data on Japanese real estate prices, population dynamics, and macroeconomic variables. Results indicate that NNs were more accurate than RF. This finding differs from that of Fonseca et al. (2023). Therefore, it is important to determine the appropriate method.