Accuracy is not always what is required for machine learning.

Many people use "weather forecasting" every day.

Weather forecasting" is a practical example of using machine learning, which is known as AI, as a solution.


In weather forecasting, it is commonplace to verbally explain in detail not only the percentage chance of precipitation, but also the strength of the wind and the timing of clear skies. When using machine learning as a solution in practice in this way, we notice the following

1. The output image is often "explanation" or "interpretation" rather than "prediction.

2. When classifying classes, only data with high probability is output, and data that cannot be classified is sometimes not output.

3. Even though machine learning is often talked about, BI tools tend to be used more in actual business operations. (Since the calculation method of AI machine learning is complex, it is difficult to understand what kind of calculation is being performed, and even if a future prediction is made using the learning model, it is difficult for those who want information based on the data to understand whether it is 100% the information they want to obtain. Therefore, BI tools tend to be used to easily calculate the original data and the processed data.)


※2.  The "classifier may only output data with a high probability of being classified and may not output data that cannot be classified." For example, if there is an image recognition system that classifies fish, frogs, and beetles into "living things in water" and "living things classified as insects," fish with fins and backbones are "living things in water" and beetles without backbones are "classified as insects. However, a frog, which has a backbone and can be immersed in water, is easily excluded from the classification category because it is an amphibian, the probability is reduced accordingly, and depending on the set probability threshold, it may not be output because it cannot be classified).


There are also cases where a machine learning system has been developed for a specific purpose, and the accuracy of the system is so high that it has no problem, but in the end, the system was not even put into practice.

This is because the calculation methods for machine learning and prediction of AI are too complex to begin with and tend to become a black box from the viewpoint of the recipient, so AI itself tends to be perceived as vague before the accuracy of machine learning. Therefore, AI itself tends to be perceived as ambiguous before the accuracy of machine learning.


However, even so, the world today tends to try to create highly accurate machine learning models whenever it hears the word "machine learning," and in fact, there is even a track record of technological advancement thanks to the pursuit of accuracy.

Even so, in practical situations, including business, there are many cases in which a superior machine learning model with higher accuracy is not used due to various practical restrictions, and something else is used instead, which suggests the difficult reality that practical needs do not always match the accuracy of machine learning. This seems to indicate the difficult reality that practical needs do not always match the accuracy of machine learning.


The above explanation indicates that users who are seeking AI in practice are not particularly interested in the process from data capture to output, or they are not anxious about AI, or they are able to communicate the flow of AI processing in a way that anyone can understand, and they tend to be able to empathize with it. If not, the provider tends to be able to explain the calculation method in a way that anyone can understand. Otherwise, if we recommend BI, which visualizes and outputs reliable data with easy-to-understand calculation methods, our clients tend to be sympathetic to our recommendations.

We provide support for the development of AI systems and the introduction of BI tools on Amazon Web Service. Please contact us for more details.

Author: Y.M

Leave a comment

All comments are moderated before being published


Contact Us

Please feel free to contact us for more information about our products and services.

Contact Us

Connect with managers and entrepreneurs

Recommended Books

Because we live in a video and virtual world, reading books is a great asset in itself. We believe it is important for us to do business with books and to share with you the books that are the DNA of our success stories, the concrete examples and abstractions and applications of those books.

Click here for a list of recommended books.

Conserve power consumption


Uuan Yanaka Documentary

Road to International Film Festivals