Choosing the right Predicitve Model

[Tal]

Hello Data Mining Experts. Need your advice  here 🙂

Never built any predictive model before  but I  have read  a lot

I have a  simple data of 28 months .

Each month has 2 variables x and  y denoting number of clicks (either 0 or 1) happened on a daily basis

I create another variable z=sum(x)/sum(y).

So having these 28 tables and the value of z in each of the 28 tables I need to forecast the value Z in the  year of 2019

What I need to understand here is :

1.will z of all these tables be enough to  predict its  value in 2019( model z=sum(x) sum(y)) ?

2. do I need to predict sum(x) and  sum(y) first if that is possible before forecasting z?

3/ What would be the right model to accomplish 1 or 1 and 2?

This might be  a silly question but I would take any advice  :). Thanks

January:

x   y

------

0  1

1  1

0  0

....

....

....

[PaigeMiller]

There's a lot of information that needs to be considered when you select a modeling technique, before you can get any reasonable answer, and if I were you, I'd find a statistician at your company or university, and consult with him or her.

Some of the many things you would need to consider (not a complete list)

1. Is the value of z seasonal or otherwise correlated in time
2. Is the value of z expected to be linearly dependent only on sum(x)/sum(y), or could there be other polynomial effects (like sum(x)/sum(y) squared, etc.)
3. Does the value of z depend on sum(x) individually and sum(y) individually?

[Tal]

Thanks for the  quick response PaigeMiller

1. don't know what u mean by seasonal

2.it will always be linearly dependent on sum(x)/sum(y)

3.Did not get that. I mean z=sum(x)/sum(y) so it depends on both variables  

No statisticians in my team or nearby 😞

[PaigeMiller · Accepted solution]

The only way to predict z for 2019 is to have predictions for sum(x) in 2019 and sum(y) in 2019. This requires a time series model, which could be seasonal, could be auto-regressive, or some other type of time series to predict the next year.

[Tal]

Thank you Paige Miller,

That is a  good start. Will look for some info on  time series models

[Tal]

Hi Paige,  so here is  my  actual data. This looks like trend  right, linear trend?

and I can use the below to predict values for 12 months

proc  forecast data=tt interval=month lead=12 out=yyy;

id date;

var sum_x sum_y; run;

I also looked it up and  saw proc  forecast with  hold,stepar,ses methods...

Would you recommend  some  of those?

date    sum_x   sum_y
----------------------------
201511 3088 7828
201512 2312 7260
201601 2415 5331
201602 2498 5411
201603 3001 6470
201604 3383 7333
201605 3709 8078
201606 4670 10331
201607 4070 9153
201608 4092 9194
201609 3780 8299
201610 3239 6871
201611 3270 7044
201612 2528 5295
201701 1831 4062
201702 1810 3832
201704 2119 4845
201705 6156 13772
201706 3784 8590
201707 3149 7437
201708 3188 7587
201709 2801 6726
201710 2702 6234
201711 2548 5746
201712 2382 5506
201801 1482 3455
201802 1412 3475
201803 1644 4031