Data Geek Challenge – government income on cars… predicted
Last week I posted my first blog in the data geek challenge (http://scn.sap.com/community/lumira/blog/2013/08/14/data-geek-challenge–government-income-on-cars) where I took a look at our governments income based on our cars… I noticed from my first chart, that predictive could probably be used on the data set. Now, it’s time to take it to the next level.
Let’s get our main data file into PA:
We “enrich” our data so that INCOME becomes a measure.
Next, we filter both on columns TYPE (Traffic tax) and PAID BY (Both). The reason is that somehow I could not find a way to “aggregate”
all the values by year (by summing the different values for TYPE) and for PAID BY we only have detailed data for 2 of our TYPES.
Now, let’s “Predict”:
I selected Triple Exponential Smoothing as that seems to be the only one that allows me to have “Year” as Period (the others require further detail like quarter or month).
when you “drag” the algorithm down towards the source data, it will “automatically” connect both parts of the analysis. Hover over the algorithm and select “Configure Properties”:
Let’s just show a trend for now:
Save and Close and run the analysis (green “play” button on the right):
looks pretty fine already, hum? Let’s do a prediction for the next 3 years. If we change the Output Mode from Trend to Forecast, we get an extra parameter “Periods to Predict”
Let’s run the analysis again:
Incredible, almost 1700 million euros on “just” Traffic taxes…
So far we’ve kept the “default” values for the algorithms variables… and it seems in this case that’s good. How about “playing” around with those… first let me change the filter on TYPE to “taxaction on car Insurance”.
this was achieved with the following values: alpha = 0.3, beta = 0.7 and gamma = 0.1 and gives us an 84% goodness of fit! You can find this data back by clicking on the “Algorithm Summary” (icon next to the chart icon):
I could go on playing with this, but I’m already depressed from these results 😉