Microbase Releases Nationwide Estimated Vacant House Rate Data by Town Unit (Open Data) for Free

Microbase Inc. has created and released for free, as open data, the 'Nationwide Estimated Vacant House Data by Town Unit,' which estimates the number of vacant houses at the town level using public open data such as the national census. The aim is to support municipalities in understanding local conditions and initiating policy discussions.

The company has been designated by Toyota City, Aichi Prefecture, as a support organization for vacant house management and utilization, and has been working on research and surveys related to vacant house management or utilization under Article 24, Paragraph 4 of the Special Measures Law for Promoting Vacant House Countermeasures. This initiative is part of the outcomes of those activities.

Background of the Initiative

Vacant houses, if left unmanaged, can deteriorate over time and negatively impact the surrounding environment through risks of collapse, hygiene issues, and visual blight. To prevent such situations, municipalities across Japan are implementing measures such as issuing guidance to dangerous or poorly managed vacant houses and registering or updating vacant house banks.

As part of these countermeasures, vacant house appearance surveys (on-site inspections) are being conducted by municipalities nationwide to understand the actual status of vacant houses. These surveys are crucial for assessing the prevalence, risk level, and management status of vacant houses. However, they require significant labor, time, and personnel for tasks such as narrowing down survey targets, creating survey maps, conducting on-site inspections across large urban areas, and maintaining registries. Given budget constraints, it is structurally difficult to conduct these surveys regularly. In practice, such appearance surveys are often conducted only once every five years, making it challenging to capture timely data in today’s context of rapidly increasing vacant houses. Moreover, by the time a property is identified as vacant, several years may have already passed, leading to advanced structural deterioration and difficulty in contacting the owner.

Objectives of This Initiative

Therefore, we considered that by utilizing open data or internal administrative data, it might be possible to easily estimate 'where' and 'to what extent' vacant houses are likely to occur. This would make it easier for municipalities to decide when and where to initiate surveys or policy actions and to prioritize their efforts.

This dataset aims to support vacant house countermeasure operations by providing material to form hypotheses about 'areas likely to have many vacant houses' at the town level, based on public statistics such as the national census. It enables a simple yet reasonably accurate understanding of micro-level volume trends prior to conducting appearance surveys or registry development.

Estimation Methodology

The data was created through the following process:

- Training Data Creation: A training dataset was constructed by combining Toyota City’s vacant house survey results (appearance survey) as the target variable with open data such as national census data (population statistics) as explanatory variables. - Model Training: Using the training data, a machine learning model was built to learn the statistical patterns associated with areas prone to vacant house occurrence. - Nationwide Estimation: The trained model was applied to national census and similar datasets to estimate and output results at the town level via machine learning.

The model was trained and tested using Toyota City’s vacant house survey data (appearance survey) with an 8:2 split. Testing confirmed a high level of accuracy, with a coefficient of determination (R²) of 0.7994.

Data Download

The data can be downloaded via the form below. Please complete the form when downloading. Your responses will be used as reference for future development.

https://forms.gle/rDoNE9Aru7z4iiwd6

For data specifications, please refer to the following document:

https://speakerdeck.com/microbaseinc/quan-guo-ting-zi-dan-wei-kong-kijia-lu-tui-ding-detaver1-dot-0detashi-yang

Data License (Usage Terms)

- License: CC BY-ND 4.0 - Commercial Use: Permitted - Redistribution or Publication of Modified/Edited Data: Not allowed - Attribution: Required. When using or citing the data, please clearly state the source as: 'Nationwide Estimated Vacant House Rate Data by Town Unit (Microbase Inc.)'

Data Positioning

Important Notes

Please understand and acknowledge the following before use:

- This data consists of estimates and is not intended to replace existing vacant house survey methods such as on-site inspections. It is designed for use in planning, prioritization, and initial policy considerations. - This is an experimental initiative, and the company assumes no liability for any damages arising from the use of this data. - As the data is still in experimental development, it may contain inaccurate estimates. Estimation accuracy tends to be lower in non-residential areas such as industrial or commercial zones and in densely populated areas, where census population statistics may not adequately explain housing conditions. - This dataset was trained and estimated using Toyota City’s vacant house survey results (primarily appearance surveys of detached houses) with national census and other open data as explanatory variables. As the definition and estimation methods differ from those used in housing and land statistics, this dataset is not aligned with official housing statistics. The estimation primarily relies on census population data and does not directly consider building deterioration, actual usage, or physical conditions (e.g., damage status). - The dataset includes a column labeled 'vacant house rate.' Ideally, vacant house rate should be calculated as 'number of vacant houses / number of housing units.' However, in this dataset, it is defined as 'estimated number of vacant houses / number of households,' which means it does not represent the standard vacant house rate and tends to show lower rates in urban residential areas. Due to reliance solely on open data and limited available information, this value is provided for reference only.

Future Developments

This open dataset has been released experimentally based on prior research outcomes. While challenges remain, we aim to release it publicly and develop future versions based on user feedback and requests.

Plans for Future Versions

This version primarily uses 2020 national census (2020 fiscal year) data. We plan to create and release an updated version incorporating the latest data when the 2025 national census (2025 fiscal year) small-area data becomes available.

Call for Collaborating Municipalities

Currently, the volume of training data is limited, and the model may not fully capture regional diversity. We are seeking collaboration from municipalities willing to provide vacant house survey results, offer advice on data utilization, or help clarify practical implementation issues for future development and releases.

We hope this open data release will encourage municipalities to use their own survey results and administrative data to achieve more precise vacant house assessments.