A.I. Powered Key Influencers Visual Explained. What and How to use it?

With August 2019 release of Power BI Desktop, Power BI team made the Key Influencer Visual General Availability and now you can find the Key Influencers even against a Measures.

What is the Key Influencer visual?

It is an A.I. powered visual introduced into the Power BI. You do NOT have to pay any extra license to use this robust feature. It just comes up with Power BI desktop for free. As per the reading, this is the first A.I. Powered visual we can find in the Power BI. Sounds like more to come 😎

This visual is capable of analyzing any variable or metric along with other variables or features in your dataset, to determine how much influence those variables made. In this post, I’ll demonstrate to you how it works with an example. Then you can understand how easy it is and how important it is for your analysis. If I elaborate more about the importance, you can use this visual to perform advanced analytics and machine learning work. When you train a Machine Learning model, feature selection is a key step which greatly impacts the performance in your model. Rather writing Python or R coding and see which set of variables give a higher impact to the train dataset target variable, you can simply use this visual, Key Influencers to drag and drop this and visually see within few seconds.

There are three main field areas in this visual. You can drag and drop a metric to Analyze place holder, this is the variable you need to analyze. In the Explain by is the place holder, you can drag and drop all other variables you need to see the influence.

As I mentioned above, from the August 2019 Power BI desktop release onward, you can use a Measure as a Metric to analyze. Usually, it is automatically upgraded into the latest version. If you haven't got the latest Power BI desktop version already download it from here. 

How it Works? 

Best way to learn something is by looking at how to use it by real-world scenario. Let's see how the Key Influencer Visual works by an example, 

Titanic Machine Learning from Disaster dataset is popular dataset in Kaggle you can use to predict the survivals based on the available train dataset. This is a good starting point for anyone who is looking for a kick-start to the Machine Learning. World-famous Titanic ship sank in 1912, 15th of April by colliding with an iceberg. During the disaster, 1502 were killed out of 2224 total passengers. 

You can go through the contest and find more information from here. In the Data, you can freely download the Training and Test Dataset. Let's have a look at the Training Dataset. 

So when you look at the dataset, you can see some of the attributes does not impact the final result at all, to determine a passenger would Be Survived or Not. Ex: PassengerID, Ticket Number. 

However, you don't require to write any lengthy Python or R code to determine which features gave more impact/influence on the target variable. 

There are two main tabs in Key influencer visual.

Steps to Do:

Download the Train.csv dataset from the Kaggle and Import into Power BI. (You can use your own dataset which you need to analyze)

In here, my metric is Survived because I need to analyze how the other attributes impact to the passenger's survival. You would notice, the datatype of the Survived is in Numeric format. Make sure to convert it into Boolean. Otherwise, it would not work. You can simply go to the Model tab and select the column, then change the datatype to Boolean, True/False. 

Then go back to the report tab, and drag and drop the Key Influencer visual into the canvas.  Then you have to configure the visual by adding fields in your dataset. 

In here we are going to analyze which fields or features are influenced to make survive any passenger. So, our metric is Survived. Which contain True/False values. True means survived and False for passengers who died. 

Next, we need to add other attributes which we might thing affect the prediction. I added Parch, Pclass, Sex, Embarked, Cabin and Age. I haven't added any field to Expanded By place holder, as our Metric is Not a measure. It is not an aggregated one. So we can't go into a more granular level than this. 

What it Explains?

First of all, I'll explain what are the components you can see in this visual. Mainly you can find two tabs. Key Influencer tab and Top segments tab. 

Key Influencer tab 

1. The dropdown basically represents the distinct values, if its categorical or else intensity if the variable is continuous. In this scenario, we have a boolean variable. Survived or Not. Set it to True, then you can see what impact to survive. 
2. In here, from the order of the top of the variable to bottom represent the influence to the Survival. Sex has more influence and then Age. Then the class. It shows First class passengers has more likelihood to be survived. 
3. At the top, Sex is Female, 3.68x times survived than sex is male. This is how it reads. 
4. You can click the circle, to visualize it in details at the right-hand side. Green color represents the influenced class and in dark gray color all other classes. In sex, we have only two classes, Male and Female. If you click the age cycle, you can see more classes in a bar chart as below. It says young passengers age below 0.92 has more influence to be survived. And it is 2.5x times than other age bins. 
   
5. The average line is calculated all the possible values for the age except less than 0.92. In other words for all the values in gray color bars. It tells in what percentage other values has a lower influence. 

We can become for a conclusion when surviving passengers, the rescue team has given more priority to female, then kids. After that first-class passenger likewise. This is the order of the features or fields in the dataset makes the impact to the result. 

Top Segments Tab 

Earlier we saw the individual variables as influencers. In this tab, from this visual, it considers as a combination of variables which influences and group them into segments.  In this Titanic dataset, it has identified three segments. These segments ranked by percentage. 

If you click one of the circles ( segment)  you can expand for details and see what are the variables contributed to that segment. 

This means, 98%  of Passengers who have not Embarked S, not 3rd class and are female. You can find more details analysis in each segment. Ex; If you select segment 3 and click > Learn more about this segment, you can find how the survival distribution affect based on the cabin. In here there are lots of missing data points for cabin details. So, we cannot take any accurate analysis from this chart. 

Summary

In this blog post, I just wanted to explain the importance of the Key Influencer visual. Because, sometimes I found my-self its not highlighted enough its Machine Learning powered capabilities in existing resources on the web. So, its encorages me to write this post to elaborate it features with practical example. Please hit the subscribe button if you found this interesting for more articles like this. Also please don't forget to share with others. Cheers! 😊