topic Moving average for predicting future values in App Development

Moving average for predicting future values

TinaH — Mon, 14 Feb 2022 15:43:32 GMT

Hello together,

I am currently having a problem regarding the use of "moving averages" to forecast values.

Basically, I have the creation month "cr_month" of tickets and the amount of tickets created per month as a dimension (grey bars). Now, I want to forecast the number of tickets for the next 12 mnths using a linear model, which is working fine via linest_m and linest_b, and the moving average (cf. screenshot).

For the moving average, currently I am using the following logic:

if ( cr_month < 202201 , count(distinct Ticket_ID) , RangeAvg(Above(count(distinct Ticket_ID),0,12)) )

So my question is: How do I estimate /forecast the number of tickets for the months 202202 - 202301? Unfortunately, the moving average curve is currently decreasing.

Thank you very much!

Best regards
Tina

Re: Moving average for predicting future values

chris_djih — Tue, 15 Feb 2022 08:21:16 GMT

1. moving average and forecasting are two fully seperate things!
as you already solved the linear variant. have a look at the other statistical forecast variants (exp, log).

Logarithmic ($1 iy your measure (count) and $2 is your date-field):

linest_m(total aggr(If($2<=today(),$1),$2),(Aggr(Log(Rowno()) * $2, $2))) * (Aggr(Log(Rowno()) * $2, $2))+ linest_b(total aggr(If($2<=today(),$1),$2), (Aggr(Log(Rowno()) * $2, $2)));

exponential ($1 iy your measure (count) and $2 is your date-field):

exp(linest_b(total log(aggr(If($2<=today(),$1),$2)),$2)) * POW( e(), linest_m(total log(aggr(If($2<=today(),$1),$2)),$2) * $2);

If you want a more detailied actual forecasting or as you mentioned areal "estimating" you have to add advanced analytic connections to use machine learning models.

Re: Moving average for predicting future values

TinaHensel — Tue, 15 Feb 2022 08:38:58 GMT

Thank you!