Forecasting/Future prediction assistance

#1
Hi guys,

I've got a bit of advertising data here and I'd like to make a model out of it which can predict future events.

So I have the amount of money spent, the number of billboards we've got and how many people we think have seen the billboard. We also have the number of walk ins we think are due to the billboard but want to know how many more walk ins we'll have if we invest more/get more exposure/get more billboards around the place.

I'd like to do this in R but I'm not sure how to do it. Could someone lead me into a direction about what I should be doing to get a decent prediction (no seasonality).

Feel free to ask me more questions regarding all of this! I'd love some input and help from people who are good at stats (unlike me).

Thanks
 

rogojel

TS Contributor
#2
hi,
could you be a bit more concrete? there are many ways to build a predictive model but I guess we would need a bit more flesh to chew.

regards
 
#3
hi,
there are many ways to build a predictive model but I guess we would need a bit more flesh to chew.
I'm just looking for a very basic model based on billboards, spend and number of people who have seen our billboard which can predict (to some extent) how many people will come into our store. I'd also like to be able to easily assess fit which is important. What other information would you need and could you recommend any models? I've tried working with just basic correlation and it's not bad, but I can only work with one variable at a time. I'd be interested in combining 2 or 3 of these into a single model and seeing how they effect everything.

Thanks

Edit: I would also like to try something a little more advanced than a multiple linear regression model. During my time studying, modelling, stochastic processes and time series weren't my strengths (in fact I barely passed those courses) so I've come to find assistance from you guys.
 
Last edited:

rogojel

TS Contributor
#5
hi,
maybe you could look at regression trees as well as multiple regression? Is your main goal to make predictions or more like getting an insight into which factors play a more important role? Models that are probably better at prediction might be difficult to explain.
How many measurements do you have?

regards
 
#6
hi,
maybe you could look at regression trees as well as multiple regression? Is your main goal to make predictions or more like getting an insight into which factors play a more important role? Models that are probably better at prediction might be difficult to explain.
How many measurements do you have?

regards
Sorry, haven't had the chance to log on in a while.
I'm looking to make predictions.
I only have the measurements that were named above. I've tried multiple linear regression analysis and that seems to work ok. When I remove an unexpected increase in activity the fit of the model decreases though (but this is a problem for another time). I'm not sure how I would implement regression trees for prediction work when the number of walk ins is continuous and numeric.
 

rogojel

TS Contributor
#7
. I'm not sure how I would implement regression trees for prediction work when the number of walk ins is continuous and numeric.
hi,
that is not a problem with regression trees, they work with continuous DVs.
How do you decide whether a model is a good predictor or not? Are you using cross-validation?

regards
 
#8
hi,
that is not a problem with regression trees, they work with continuous DVs.
How do you decide whether a model is a good predictor or not? Are you using cross-validation?

regards
Hi rogojel, sorry other things have been keeping me busy so I've been unable to come back and reply to this thread!
If you're still around (and interested), how would you go about implementing regression trees in this case? I'm not that interesting in using cross-validation.
Thanks
 

rogojel

TS Contributor
#9
hi,
there are several packages in R implementing trees. I do not know about other stats programs but I kinda assume the bigger ones should have the method.

You will definitely need some measure of prediction performance - if you are not using CV, then what?

regards
 

hlsmith

Less is more. Stay pure. Stay poor.
#10
I believe earlier rogojel was asking how many observations do you have. You have listed your variable titles, but can you provide data for say the first 10 rows of your spreadsheet, so we can better understand your dataset. Also, can people be exposed to multiple marketing components?


Also, if you have built a basic multiple regression, can you post that code so we can see how you are trying to layout these variables?


Thanks.
 
#11
Hi guys, I will look into using trees in R. I wasn't completely sure how to measure fit. I was hoping the output in R could assist me and tell me if the fit was good. I wasn't interested in cross validation because I'm not too familiar with it but will also research it further to try and understand it more.

I've attached a picture of what my data looks like: https://i.imgur.com/dnhACIs.png
The data above is the spend, boards, viewers and walkins we've had per month.
No, it would not be possible for people to be exposed to multiple marketing components.

The code I used for multiple linear regression was something like:
model = lm(Walk Ins ~ Spend + Billboards + Viewers)