How Can Artificial Intelligence and Power BI Derive Smarter Insights?!

Power BI is a business analytics tool that helps businesses make better decisions. In recent times, Power BI's AI capabilities have been discussed a lot. This blog explains significant AI-driven features like Q&A functionality, predictive analysis, and AI visuals.

1. AI-Generated Graphics

Power BI provides two essential visuals powered by artificial intelligence that assist users in gaining insights from data with little effort.

a)   Key Influencers Visual: 

This tool helps to understand how different metrics affect a chosen statistic in the data. In other words, it analyses the data and determines which are the most impactful individual contributing factors and what their significance is. It clearly explains the distribution and cause of the values increasing or decreasing.

Let’s see how the key influencers' visual is created.

• Under Build Visual on the Visualizations pane, select the Key Influencers icon.

• Move the metric to investigate into the Analyze field.

• Using a nutrition score as an example, we can see that the status of the nutrition score can be explained by using a key analyzer, high-calorie diet, hemoglobin diet, breakfast meal, afternoon snack, cookies, vegetables, supper, and lunch meal.

 Image_1:

• As shown in image_1, Vegetables are the top influencers and they impact positively with a Nutrition score classified as “Healthy”. It increases the likelihood by 3.92 times.

• The bar chart on the right shows that the percentage of “Healthy” nutrition scores is higher when vegetables are part of the diet.

• The average percentage without vegetables in a dotted line is at 22.33%

Image_2:

• This image_2 displays that having a low Hgb (Hemoglobin) diet is the top negative influencer contributing to an “Unhealthy” nutrition score of 6.86 times.

• Absence of meals and snacks contributes to an “Unhealthy” nutrition score.

• The bar chart on the right shows that a low Hgb diet is the significant factor for the higher percentage of “Unhealthy” nutrition scores.

• The average percentage without Hgb diet is indicated by a dotted line at 11.83%

Image_3:

• Three key segments were identified based on the percentage of Nutrition scores rated as “Healthy”.

• Segments 1 and 2 with population count of 68 and 34 respectively have a higher percentage of a healthy nutritional score of 100%.

• Segment 3 has a lower percentage of healthy nutritional scores (79.1%) with a population count of 43.

  
  

Image_4:

• Image_4 identified 4 segments based on a percentage of Nutrition Scores classified as “Unhealthy” and given their population count.

• Segment 1 has the highest percentage of bad nutrition scores, despite being the smallest segment of the 18 population count.

• Segment 2 is notable because of its high percentage of unhealthy scores (87.9%) and moderate population size (33), making it an important group for focused nutritional interventions.

• Despite being the largest group, Segment 4 had the lowest percentage of unhealthy scores among the four, indicating that different intervention tactics are required than in the other segments.

Image_5:

• For an example, let’s see the characteristics of one of the top segments for “Unhealthy Nutritional Score” from the image_5.

• In Segment 2, 87.9% of the Nutrition Score is “Unhealthy”. Image_5 suggests that the absence of snacks may correlate with an unhealthy nutrition score.

b) Decomposition Tree Visual: The Decomposition Tree visual is used to drill down into multiple dimensions of the data and investigate the root cause of a certain condition by selecting the highest or lowest values. The visualization requires two types of input:

Analyze – the metric you would like to analyze. It must be a measure or an aggregate.

Explain – one or more dimensions you would like to drill down into.

After dragging the measure into the field, the visual updates showcase the aggregated measure. In the example below, we are visualizing the count of patients with specific medical conditions and root causes.

The analysis can work in two ways depending on your preferences. Using the supply chain sample again, the default behavior is as follows:

• High Value: Looks at all fields and determines which one to drill into to get the highest value of the measure being analyzed.

• Low Value: The lowest value of the measure is analyzed after examining all fields.

2. Q&A Feature

The Q&A feature in Power BI explores data in our own words, as a result, it gives the answers to the questions that are asked in natural language. Q&A is available on dashboards in the Power BI service and can be added to reports in the Power BI service or Power BI Desktop.

Q&A can be added to your reports in a couple of ways.

• Select from the viz pane as shown below.

• Double-clicking into the background of reports

• Use the insert Tab and see Q&A, there are other AI visuals built into Power BI here along with Q&A.

• There's even a Q&A option to select from the Buttons tab if you would like to create a smaller version of this in your report.

 How It Operates:

The Q&A feature allows users to ask natural language queries for example "total systolic BP change over time?" as shown in the image below. Then, depending on the query, Power BI selects the suitable visual automatically.

Suggestions: An additional option is to select a question from a highlighted list of ideas.

Convert icon: Select this icon to convert a Q&A visual into a regular visual with the edit view option.

Gear Icon: The Gear Icon is utilized to access the Q&A Tooling pane, which gives us the ability to adjust the natural language engine. Only in the Edit view is the gear icon accessible.

Q&A button in a report: Add a Q&A button to reports and open the Power BI Q&A Explorer window. Visuals created using the Q&A button are for exploration only, they can't be saved to the report or dashboard and the Q&A button is only available in Reading mode.

 

Limitations: At the moment, Power BI can only support English natural language queries.

3. Predictive Analysis with Power BI

Predictive analysis in Power BI uses machine learning to estimate future trends using historical data. This is especially useful for businesses seeking to make proactive decisions.

Forecasting: Power BI has built-in forecasting tools that predict future data trends based on past patterns. For example, you can forecast future revenues, customer growth, and inventory needs. Power BI's forecasting model may be adjusted by changing parameters like forecast length, confidence intervals, and seasonality. Forecasting with Power BI may aid with informed decision-making, resource management, and risk minimization.

AI Insights: Power BI's AI Insights feature works with Azure Machine Learning, allowing users to apply machine learning models throughout the platform. Users can connect to pre-trained models or construct new models in Azure, to use with Power BI data sets.

Text analytics: The AI Insights tool can be used to analyze text data. This feature can be used to extract insights from unstructured data and detect trends and sentiments.

Anomaly Detection: Anomaly is defined as “an unexpected change within the data patterns, or an event that does not conform to the expected data pattern”. In short, Anomalies are abnormalities. Anomaly detection aims at finding unexpected or rare events in time series data streams.

Why should we detect Anomalies?

The anomalies indicate major variations and deviations from the expected trend. The variations might be due to errors in data collection or technical glitches.

Methods for Finding Anomalies using Power BI:

Using the Gestational Diabetes Dataset, abnormalities can be found by taking the two columns named "Date of form signed" and "25-OHD result date". 

Create a line graph with "Date of form signed" on the x-axis and "25- OHD Result date" on the y-axis. Below is the reference picture.

To detect Anomalies, go to the Analytics Pane and turn on “Find Anomalies”. Change the sensitivity according to your requirements.

If there is an increase in sensitivity, the algorithm is more sensitive to changes in the data. Even a slight deviation is marked as an anomaly.

Now, note the change in your line chart. The anomalies are marked with a marker that can be formatted. Here, one anomaly has been detected.

Red colored square is an anomaly detected.

Limitations:

It is only supported for line charts with time series data on one axis and more than three data points on the other axis.

The line chart visual does not support secondary values, multiple values, or legends.

Conclusion:

The combination of Power BI with AI and Machine Learning represents a dramatic shift in business analytics. It enables enterprises to transition from static reports to proactive tools to ensure success. The combination of these technologies will have a huge impact. Businesses can now use Power BI to not just analyze the present but also predict and change their futures.