Скачать или смотреть Two Approaches to Seasonal Forecasting with MicroStrategy

This demo shows how MicroStrategy can build forecasting models that incorporate seasonality into the predictions using two algorithms: Regression (native) and ARIMA (using R).

TURN ON CAPTIONS TO SEE THIS COMMENTARY:

00:02
Let's use MicroStrategy's native machine learning algorithms to forecast Monthly Revenue

00:06
In Developer, right-mouse-click on the Revenue metric and insert a new Training Metric

00:10
"Training Metrics" apply machine learning algorithms to the data on this report

00:13
The ML algorithm builds a "Predictive Metric" to forecast Revenue into the future

00:17
Fortunately, Developer's Training Metric Wizard does it all in 3 easy steps!

00:20
The first screen captures the Type of Analysis to perform, which depends on what we're trying to accomplish

00:23
There are three ML algorithms that forecast numbers

00:25
Two ML algorithms that predict outcomes

00:27
as well as a K-Means Clustering algorithm

00:28
and an Association Rules algorithm

00:29
We're forecasting numeric data so let's choose Linear Regression and click next

00:31
This second screen identifies the data we'll use to build the model

00:34
Revenue is the "Dependent Metric" representing what we're trying to forecast

00:36
And we have two "Independent Metrics", Month Index will capture the general trend of the data and Month of Year will capture seasonality

00:40
The third and final screen is where we specify the name, location and other features of our Predictive Metric

00:46
Click Finish to execute the report and create our Predictive Model to forecast Revenue into the future

00:53
Let's do some rearranging and go into graph mode to see a chart of the forecast

01:04
Because this is a Predictive Metric, we can right mouse click on it to see additional details about the predictive model

01:09
We can view the regression equation that defines the model along with lots of other model details

01:12
We even export the PMML representation of the model that can be used to the deploy this model to another system for scoring

01:19
The regression forecast has a sawtooth pattern that repeats for each year following the generally increasing upward trend

01:25
Next, let’s use a different ML algorithm for forecasting, one included with MicroStrategy's R Integration Pack, the "Off-the-Shelf" R script for ARIMA

01:30
Because "Near money is dear money”, Time Series algorithms like ARIMA give more weight to more recent data points

01:38
Simply copy the metric expression to deploy this R analytic with MicroStrategy

01:48
We can use any metric editor to insert our ARIMA analytic

01:56
We insert a new metric that will we will call ARIMA and paste in the metric expression

02:03
We only need to replace the target input variable with our revenue metric that we want to forecast and click OK

02:17
We see the new ARIMA forecast seems to track more closely of the more recent data points as compared to the regression forecast, which is significantly lower because it gives equal weight all three years of data