First, let's start with the content of the announcement made by the business award-winning team Gurunavi DA. The team of Gurunavi's two data administrators (Mr. Hidetomo Miyazawa and Mr. Masanori Nakajima) raised the theme of "You can do it even if you have zero work experience in analysis! We will challenge restaurant waste loss with 'Oracle Cloud' machine learning." .
Mr. Hidetomo Miyazawa of Team Gurunavi DA120 tons of food is wasted in the Japanese restaurant industry every year. Gurunavi DA considers one of the reasons for this to be "the downswing in actual numbers of visitors and orders compared to the expected number of visitors and orders", and worked on how high prediction accuracy can be achieved with ADWC and Oracle Analytics Cloud (OAC).
The data used are open data such as calendars, weather information, and event information, as well as sample data collected and processed from actual stores. First, let the machine learning model of OAC learn the performance data of 2017 and create both prediction models for “number of visits” and “number of orders”. Next, based on the weather data of 2018, etc., we predicted annual store visits and orders, and evaluated the results by comparing them with actual results.
Mr. Miyazawa says, ``We are amateurs when it comes to data analysis, and have little knowledge of deep formulas and models. If you open it, the result will surprise you. First of all, we were able to reproduce not only yearly fluctuations in the number of visitors but also weekly fluctuations almost accurately.
Predicted (blue) and actual (yellow) visitor numbers. The two lines almost overlapped, and we were able to predict with high accuracyOn the other hand, we prepared two menus, menu A without seasonality and menu B with seasonality, for the number of orders. Although menu A was able to predict with high accuracy, menu B was not a good result. When we investigated the cause, we found that the temperature data was not being used correctly. At first, three data, minimum temperature, maximum temperature, and average temperature, were input, but the problem was solved by narrowing it down to one. “Regardless of whether the food is seasonal or not, we were able to achieve satisfactory accuracy throughout the year,” says Mr. Miyazawa.
Forecast (blue) and actual (green) for Menu B (with seasonality). The first model (left) had a large discrepancy, but the model (right), which was adjusted to handle temperature data correctly, was able to make highly accurate predictionsMr. It was justified enough," he said. At the same time, he said, ``In order to increase the accuracy and actually have it widely used, it is necessary to accumulate actual data of various restaurants and more detailed data such as rainfall and wind speed forecasts.'' showed room for
Mr. Fumihiko Yamada (right) of Tokio Marine & Nichido Systems Co., Ltd., who is also the leader of the JDMC Engineers' Association, presented the Gurunavi DA team, which received the Engineer Award, with a trophy and a gift. On the left of the photo is Mr. Masanori Nakajima, who worked behind the scenes at Gurunavi Co., Ltd.