Return to page


Takenaka Corp.: Prediction of occupational accidents in the workplace using AI

Acordocerto Logo Acordocerto Logo

AI model creation to deployment time significantly reduced



Improved worksite safety



Improved efficiency of skilled workers assignment


Takenaka Corp, headquartered in Osaka Prefecture, has been involved in the construction of numerous buildings as a major comprehensive construction company since its establishment in 1610, contributing to the social development of Japan. While the labor environment in the construction industry relies heavily on manual labor and has not made significant progress in automation, it faces the harsh reality of a decreasing number of skilled workers at the construction sites. Takenaka has been actively utilizing AI to collect various data from workplaces nationwide and achieve optimization of QCDSE (Quality, Cost, Delivery, Safety, and Environment) in the workplaces.

Construction digital platform development and data utilization

With the aim of promoting operational efficiency, Takenaka has been working to collect, analyze, and visualize data generated at each site in its own business processes (proposal, contract, design, construction, maintenance). Been working on it for several years now. After the collected data is stored in the data lake, models corresponding to AI projects are developed on data utilization platforms such as BI and AI that coexist on the same digital platform.

In the past, the idea of outsourcing the model development work was also considered, but there was a problem that one model development cost hundreds of thousands of dollars.

There was also the uncertainty that the model we created might not be applied to production in the end.


The challenges faced by the company's Analytics Team

Before introducing's products are as follows:

  • The Digital Department members (a few individuals) were constantly handling 4 to 5 internal projects each, making them extremely busy and unable to meet all the AI requirements within the company.
  • To improve the accuracy of AI model development and gain acceptance from the field regarding prediction results, it was necessary to distribute the program, receive feedback, and iterate through the cycle multiple times for enhancements.
  • In the past, they had introduced another AutoML product. However, when distributing exported programs for each use case, costs were incurred, becoming a hindrance to the promotion.

By adopting's AutoML platform, Takenaka is able to efficiently handle multiple analytical projects with a limited team and achieve effective decision-making and systematization of rapid AI applications.

Currently, Takenaka has deployed around 10 AI projects in production environments, including demand forecasting to determine if sales activity reports will lead to actual project closures and evaluation analysis to analyze customer surveys for business improvement. AI is also being utilized in various applications such as image recognition and text analysis at work sites. Among them, there is a notable use case that not only improves internal operations but also considers worker safety management, which is "Prediction of occupational accidents in the workplace."

Ideally, accidents should be eradicated at work sites. However, there are cases where unforeseen accidents occur due to the nature of the business, and once an accident happens, it has significant impacts not only on the company but also as a social issue.

As the Digital Division, they have developed a solution to minimize accidents at work sites. In addition to relying on the expertise of experienced and skilled veteran workers for safety management, they utilize data from past accident cases and the corresponding work details and have developed an AI model that predicts the types of accident that are likely to occur based on the specific work activities of the day. This solution proactively alerts the work site by notifying them of the predicted results in advance. By incorporating historical data and leveraging AI, we aim to prevent occupational accidents and enhance safety measures at work sites.

The accident prediction results generated by an AI model based on factors such as the type of work site, building structure, nature of work, and past accident data are shared among site supervisors and collaborating companies. This information is then utilized during morning meetings to promote awareness of appropriate accident avoidance measures and strengthen on-site inspections. The predicted accident data, along with the daily work data, is imported into the server of a familiar business intelligence (BI) tool during the night. Users can then access the data through tablets or smartphones, where screens are displayed in order of high accident frequency for their review.

After the implementation of DriverlessAI system, Takenaka was able to achieve the most significant impact by reducing accidents in the actual work site and decreasing injuries and fatalities among workers. Furthermore, the inefficiency in manpower allocation that previously required constantly assigning skilled construction managers to the site has been improved, leading to improvements in the working environment.

quotation mark

It is worth noting that there is a demand for higher visibility tools in the field compared to existing visualization (BI) tools. Therefore, we are currently exploring the integration of digital twin technology with the results of DriverlessAI to validate a mechanism that combines 3D data of the entire workplace and contributes to enhancing safety measures.”

Mr. Nakagawa


Building flexible predictive models utilizing H2O

Takenaka pointed out that the key factor in adopting is "acceleration of deployment". Mr. Nakagawa from the Digital Division mentioned that H2O Driverless AI provides a user interface and features that allow analysts to work without interruption. He also expressed a strong belief that the rich algorithms and data processing libraries called "recipes" provided together (available on GitHub) have the potential to serve as a receptacle for any future AI projects.

In addition, Mr. Naka in the same department stated, "Due to the diverse deployment patterns, we can operate the model not only online but also in offline environments such as workplaces, which has expanded the scope of AI utilization."

In fact, the systemization and application patterns of predictive models implemented by Takenaka Corporation have achieved efficiency through the appropriate selection of architecture for each scenario.

Systemization patterns for predictive models

  1. During the model evaluation phase, users are provided with an environment that allows them to quickly obtain results using the REST API functionality of Driverless AI. They can perform multiple validations by changing the input data.
  2. A scoring environment using the MOJO (Model Optimized Java Objects) package is built on containers or cloud servers. Results can be obtained from existing systems via APIs.
  3. The scoring environment using the MOJO package is directly implemented within the system.

The flexibility of these systemization patterns is a unique advantage of, enabling the implementation that aligns with various utilization patterns of Takenaka.


The Future of Digital Division Achieved with

H2O Driverless AI not only automates the selection of appropriate algorithms and feature engineering for AI model creation, but it has also been proven to deliver equal or better model accuracy compared to models manually created by data scientists.

In the future, we plan to further leverage GPU processing and actively explore applications in various areas, including:

Improving the efficiency of daily management ‘instructions by sharing construction status with wearable cameras’, ‘Improving efficiency in creating construction records through voice recordings and on-site videos’ or ’ Prompt instructions for swift responses based on on-site videos and sensor data from experts in case of trouble’.

Takenaka is actively considering the implementation of these technologies in order to enhance operational efficiency in these respective fields.

About is recognized as a global visionary leader in the fields of automated machine learning (AutoML), time series forecasting, and explainable responsible AI, enabling the democratization of AI. With over 20,000 organizations worldwide, is a trusted AI partner. By leveraging our platform, H2O AI Cloud, businesses, government agencies, non-profit organizations, and academic institutions can utilize AI for manufacturing, operations, and innovation, accelerating responsible innovation and expanding the possibilities of artificial intelligence.

(*) was selected as a leader in the visionary segment in the 2023 Gartner® Magic Quadrant™ for Cloud AI Developer Services.