6 min to read
For more effective use of AI: XAI & Prescription Analysis
We will look at the algorithms of deep learning artificial intelligence and explain how to correctly analyze and utilize the collected data.
For more effective use of AI: XAI & Prescription Analysis
● Intro
The use of AI using deep learning technology has led to rapid development of artificial intelligence technology and an increase in market size. However, the problem is that the algorithm of deep learning-based artificial intelligence is in a black-box (internal process unknown format) structure, so the reason for the derivation results is not easily understood.
In this posting, we will discuss how DS2.AI approaches these internal algorithms on two topics: explainable AI (XAI), and the steps of data analysis.
● Why do we need the description of the process of getting results?
We do not all wonder and act about the various phenomena that happen around us. We all sleep and wake up, breathe, watch and listen, but we rarely seriously think about why all that happens and what’s the main factor. However, it is a necessary process when using artificial intelligence.
First of all, The fields that require AI are diversifying, and advancement is inevitable to handle many tasks. However, if developers can't identify the algorithm process, they have to go through a direct verification process for all variables, which will require a lot of time and cost. The coverage of AI won't be increased easily.
Also, even if you accidentally developed an AI with high accuracy, the probability of use will be very low if you fail to explain the derivation process. Let’s think of it with a simple story. Let’s suppose that in the future, AI has melted into real life and measured the potential for a certain person to cause a crime like any movie. What if I was informed one day that I was identified as a potential criminal? First, I will ask why I was pointed out and what factors affected me. If you hear the answer, “It’s just chosen by AI’s complicated algorithm to identify the possibility of crime,” it would not be accepted as a reasonable and justifiable reason. Trust in AI itself cannot be secured either.
Thus, AI can only be used as a meaningful artificial intelligence model if it is possible to explain how it works and processes processing variables in all processes of developing and introducing AI.
(Even with advanced AI, if there is no explanation of the process, the user's curiosity will only increase.)
● XAI of DS2.ai
We, DSLAB Global, have developed to show how data and variables are used numerically and schematically when the AI model creation project is completed.
The project to introduce is related to ‘bank deposit prediction’. When the project is completed, the reliability of the model is briefly shown through accuracy and error rate, and detailed information can be checked through the menus on the right (red box).
■ Feature Importance : It shows the data with a high weight used in the process of deriving the corresponding result. In the current ‘Bank Deposit Prediction’ project, it can be seen that ‘Quarterly Employment Change’ was selected as the most important variable.
■ Kappa-Coef (top) & Loss (bottom): The metrics on the graph can help you determine the reliability of your model. It gives you an idea of how reliable the model itself is and how much it can actually be used.
In addition to the representative elements introduced, DS2.AI proceeds with the process of analyzing and utilizing the derived results after confirming the degree of utilization and reliability of precise data.
● Prescription analysis based on reasonable data
After creating an environment to understand the algorithm’s operation process and major variables, we aimed to create an algorithm that could provide customers with a suitable guide as well as data analysis with a little more greed.
First of all, let me briefly explain the 4 stages of data analysis.
1) Explanatory analysis: Explain past or present facts by various standards.
2) Diagnostic analysis: Use data to determine the cause of a particular problem.
3) Predictive analysis: Predict unknown facts, such as future situations.
4) Prescription Analysis: Guide specific actions beyond the stage of analysis.
There are four stages of analysis.
(Data analysis type step by step diagram. It is difficult as it goes from the right to the top, but high value can be created)
Data analysis type step by step diagram. It is difficult as it goes from the right to the top, but high value can be created
(Diagnostic Analysis of the 'Bank Deposit Forecast' Project)
Through these graphs, we can figure out how much influence a variable has when it has a certain value.
By using this, it is possible to make a simple prediction that replication is possible and that certain variables will have to be adjusted to produce meaningfully different results. However, simple predictions are not enough to yield meaningful results.
(① "prescription analysis" of 'Bank Deposit Prediction' project analysis)
(② Selection of additional variables for 'Bank Deposit Prediction' project analysis)
DS2.AI informs the customer of the task to be performed in consideration of the variables that can have a direct impact as shown in image ①.
If you increase the number of employees who manage customers by 6.5%, increase the frequency of contact with customers by 5%, and finally contact them in May, you provide a guide that increases the chances of customers using the bank by about 76.77% in May.
In addition, as shown in image ②, it helps management make decisions with as little risk as possible by telling them what additional variables they need to keep in mind.
Based on the bank deposit prediction project, in addition to the three pre-selected variables It shows that the ‘Consumer Confidence Index’ and the ‘Consumer Price Index’ can have an effect of about 10% in total.
Summary
In today’s post, we looked at the necessity of XAI and prescriptive analysis and the algorithm analysis process of DS2.ai.
As AI is attracting attention as it will be dedicated to many tasks performed by humans in the future, accurate and reasonable AI model creation and data analysis must be accompanied.
DSLAB Global promises to be reborn as a company that provides reliable information, and we will see you later with another topic.
Take a look at the Click AI service that provides reasonable XAI and prescriptive analysis of DS2.AI and a company case that has been successfully applied to the business!
References : Should AI Explain itself? 4 cases about data analysis