Good morning! Here we are again in our curious corner of BestInMicro!
A place to share stories, experiences, learnings, secrets, and techniques from the day-to-day life of a MicroStrategy consultant.
This week I bring a topic that has been in MicroStrategy for many years, but few people know about it and companies spend thousands of dollars to implement it in an outsourced way. Sometimes it may be justified, as this is an out-of-the-box solution, but believe me, if it were better known, it would be used a lot. And yes, you are looking at it right, we are talking about machine learning techniques integrated with MicroStrategy! So, without further delay, the topic of the week: Training Metrics.
After the introduction, I will highlight something. These models correspond more to the concept of data mining than to that of machine learning, but one thing does not take away from the other. What I will try to explain in this article is a quick introduction to the different predictive models and how to use them with MicroStrategy, without the need to have a data scientist or someone who knows R or Python to predict behaviors.
For people who are used to classic BI, what we usually see and analyze is the past, the results obtained to understand what happened, and to be able to make better decisions. This is called descriptive analytics.
Today, what we will do is implement models that allow me to predict, as accurately as possible, what the results will be using statistical techniques. This type of analysis is what we call Predictive Analytics. Below are the types of questions answered by each of these analyses:
Now that we have positioned ourselves a little on the purpose of these models, I will briefly tell you what types of models there are. The point of all this is that we can create indicators to help us predict future results, but not all models behave in the same way.
Forecasting is a type of analysis used to predict or estimate the future value of variables such as sales, profits and stock control, etc., always based on past data. Regression analysis and time-series analyses are the most common to see and the simplest examples.
In this type of analysis, there is an independent variable (generally, time), and another dependent variable (on the Y-axis).
Basically, an attempt is made to generate a relationship between all the factors that affect the relationship between the two variables, and an equation is created which, when a value of the independent variable is applied to it, returns the estimated value of the dependent variable as accurately as possible. The more information we have, and the fewer factors that influence our scenario, the more accurate the analysis will be.
There are different types of formulas, such as linear regression (f(x) = mx +b), exponential regression (f(x) = mx+b), logarithmic regression (f(x) = ln x)….. here are more details about it, but depending on each situation and the factors affecting the scenario being analyzed, each formula will be better adapted, minimizing the margin of error.
According to the official definition: Classification is a technique used to assign objects to different groups and predict how to classify new data points by identifying underlying data characteristics. I finished writing it and it sounds like basic Chinese, but with an example, it is easy to understand: Based on different rules, I can assign an object to different groups. It is like an Akinator, but giant… using the If then else conditions we group new objects, which allows us to make decisions regarding their resolution.
-Is it alive? -Yes
Is it green? -No
-Does Wow? -Yes!
-It is a surprised person!
Besides being a lousy joke (sorry, I couldn’t help it), it gives us the guideline that we have to generate good rules to know how to classify and predict correctly… We will see it in the next part of the article.
These technique groups elements so that members of a group are more similar to each other than to members of other groups. The degree of association between two elements is maximum if they belong to the same group and minimum otherwise. Once groups have been created, new items can usually be grouped into one of the existing groups.
This type of analysis is very typical, for example, in marketing campaigns. If there are many people of a certain age range, gender, and region, they will look to sell a specific product, whereas, if the age range changes, they will look for another product that is more attractive to this new group or cluster.
Association rules detect patterns, correlations, or relationships between different elements or sets of elements. They help to identify elements that usually appear together and have a high degree of concurrence with each other.
For example, if a person buys a cell phone and a protective case, likely, he will also want to buy a tempered glass… or if he buys coal and chimichurri in a supermarket, he will also want a good piece of meat and wine for his barbecue…
Basically, it is to identify patterns and behaviors that are constantly repeated so that, when detected again, it can be predicted how it will end.
Data Miningdoes not indicate the value of patterns: You need to have a good understanding of the data and the business context to interpret the value of the trends and patterns discovered.
While it is not required to understand all the intricacies of statistical and mathematical techniques related to data mining, it is of extreme importance to understand the tools and techniques of data mining and their functionality, because if they are not applied correctly it can lead to very wrong results.
It is of utmost importance to have a verification phase. Data mining identifies patterns and trends based on sample data. These patterns and trends should be checked against a real-world data set to ensure that the results are reliable.
The quality of the data affects the outcome: data mining is directly affected by the quality of the data. The data must be clean and consistent across all data sources.
Data mining may not identify cause and effect relationships: While it may make it easier to identify certain predictive patterns, those patterns may not identify a cause and effect relationship.
These are tools that MicroStrategy has had for many years, but as the world advances and with the massification of new technologies, computational capacity, and statistical models, it has been migrating towards new models made with Python, for example, a new tool that MicroStrategy is adapting to with new versions and functionalities.
Anyway, I think it is still a super exploitable functionality and does not require a data scientist to take advantage of it.
Anyway, we will soon do some practice with these models, and let me know if you want me to focus on any particular model for the example, or even if you have and want to share a dataset that you want us to analyze together.
See you next week!
11.711-Data Mining and Predictive Analytics Manual
James Oliver Vídeo – MicroStrategy Training Metrics Tutorial – YouTube
Online Training Video by MicroRooster – MicroStrategy – Data Mining & Predictive Analytics – Online Training Video by MicroRooster – YouTube
* Tu información será utilizada exclusivamente para contactarte en relación al Business Data Master. No hacemos spam ni compartimos datos con terceros.
