Search results

The success of a neural network depends on, among others, an architecture that is appropriate for the task at hand. This study attempts to identify an optimal architecture of a neural network in the context of property valuation, and aims to test the ability of connecting related neural networks to reduce the property valuation error.

This study explores efficient network architectures to estimate land and house prices in Seoul, South Korea. The input is structured data, and the embedding technique is used to process high-cardinality categorical variables.

The shared architecture of a network for simultaneous estimation of both land and houses was revealed to be the best performing network. Through weight sharing between relevant layers in networks, the root-mean-square error (RMSE) for land price estimation was reduced significantly, from 0.55–0.68 using the baseline architecture, to 0.44–0.47 using the shared architecture.

The study results are expected to encourage active investigation of efficient architectures by using domain knowledge, and to promote interest in using structured data, which is still the dominant type in most industries.

It is important to forecast local trading volumes as well as global trading volumes because the real estate market is always characterized as a localized market. The house trading volume at the local level is forecast through appropriate models to enhance the predictive accuracy.

Four representative housing submarkets in South Korea are selected, and their trading volumes are forecast. A well-established time-series model and a deep learning algorithm are employed: the autoregressive integrated moving average (ARIMA) model and the recurrent neural network (RNN), respectively. The trading volumes in adjacent areas are utilized as covariates, and an ensemble prediction is applied additionally to improve the model performance.

The results indicate no significant difference in prediction performance between the ARIMA model and the RNN, which can be attributed to the insufficient amount of data used. It is discovered that the spillover effects of trading volumes across the study areas can be exploited to improve the predictive accuracy, and that the diversity of the predicted values from the candidate models can be used to increase the forecasting accuracy further.

Whereas property prices have been investigated extensively, the discussion on forecasting trading activity of properties is limited in the literature. The results of this study are expected to promote more interest in adopting a local perspective and using a diversity of predicted values when forecasting house trading volumes.

Sampling taxpayers for audits has always been a major concern for policymakers of tax administration. The purpose of this study is to propose a systematic method to select a small number of taxpayers with a high probability of tax fraud.

An efficient sampling method for taxpayers for an audit is investigated in the context of a property acquisition tax. An autoencoder, a popular unsupervised learning algorithm, is applied to 2,228 tax returns, and reconstruction errors are calculated to determine the probability of tax deficiencies for each return. The reasonableness of the estimated reconstruction errors is verified using the Apriori algorithm, a well-known marketing tool for identifying patterns in purchased item sets.

The sorted reconstruction scores are reasonably consistent with actual fraudulent/non-fraudulent cases, indicating that the reconstruction errors can be utilized to select suspected taxpayers for an audit in a cost-effective manner.

The proposed deep learning-based approach is expected to be applied in a real-world tax administration, promoting voluntary compliance of taxpayers, and reinforcing the self-assessing acquisition tax system.

Most prior attempts at real estate valuation have focused on the use of metadata such as size and property age, neglecting the fact that the building workmanship in the construction of a house is also a key factor for the estimation of house prices. Building workmanship, such as exterior walls and floor tiling correspond to the visual attributes of a house, and it is difficult to capture and evaluate such attributes efficiently through classical models like regression analysis. Deep learning approach is taken in the valuation process to utilize this visual information.

The authors propose a two-input neural network comprising a multilayer perceptron and a convolutional neural network that can utilize both metadata and the visual information from images of the front view of the house.

The authors applied the two-input neural network to Guri City in Gyeonggi Province, South Korea, as a case study and found that the accuracy of house price estimations can be improved by employing image information along with metadata.

Few studies considered the impact of the building workmanship in the valuation process. The authors revealed that it is useful to use both photographs and metadata for enhancing the accuracy of house price estimation.

Unstructured data such as images have defied usage in property valuation for a long time. Instead, structured data in tabular format are commonly employed to estimate property prices. This study attempts to quantify the shape of land lots and uses the resultant output as an input variable for subsequent land valuation models.

Imagery data containing land lot shapes are fed into a convolutional neural network, and the shape of land lots is classified into two categories, regular and irregular-shaped. Then, the intermediate output (regularity score) is utilized in four downstream models to estimate land prices: random forest, gradient boosting, support vector machine and regression models.

Quantification of the land lot shapes and their exploitation in valuation led to an improvement in the predictive accuracy for all subsequent models.

The study findings are expected to promote the adoption of elusive price determinants such as the shape of a land lot, appearance of a house and the landscape of a neighborhood in property appraisal practices.

Prior studies on the application of deep-learning techniques have focused on enhancing computation algorithms. However, the amount of data is also a key element when attempting to achieve a goal using a quantitative approach, which is often underestimated in practice. The problem of sparse sales data is well known in the valuation of commercial properties. This study aims to expand the limited data available to exploit the capability inherent in deep learning techniques.

The deep learning approach is used. Seoul, the capital of South Korea is selected as a case study area. Second, data augmentation is performed for properties with low trade volume in the market using a variational autoencoder (VAE), which is a generative deep learning technique. Third, the generated samples are added into the original dataset of commercial properties to alleviate data insufficiency. Finally, the accuracy of the price estimation is analyzed for the original and augmented datasets to assess the model performance.

The results using the sales datasets of commercial properties in Seoul, South Korea as a case study show that the augmented dataset by a VAE consistently shows higher accuracy of price estimation for all 30 trials, and the capabilities inherent in deep learning techniques can be fully exploited, promoting the rapid adoption of artificial intelligence skills in the real estate industry.

Although deep learning-based algorithms are gaining popularity, they are likely to show limited performance when data are insufficient. This study suggests an alternative approach to overcome the lack of data problem in property valuation.