Forecasting Active Users?

r
predictive_model
featureselection
forecasting
python

#1

Hi everyone,

I am new to the forum so please forgive me if this question has been previously asked!

I have been collecting data on the number of users that have used my app on a daily basis for the past 6-7 months. The data is in the following format:

date, users
2017-01-01, 450
2017-01-02, 535
2017-01-03, 288

and so on.

Now what I would like to do is essentially use this data to predict what the value for users will be each date in the future. What would be the best way to approach this problem? It seems to me that this is similar to a “standard” sales forecast, but I am not really sure. Is time series appropriate here? Would a machine learning algorithm be useful? Would i need to collect more data of the most relevant factors that can influence the number of users?

Many thanks in advance!


#2

@JuanSGiraldo

First of all - welcome here!

In order to forecast the number of users, you need to ask yourself the following question:

Will number of users to your app depend only on time or other factors as well?

If you need a simple time based model, time series is the way to go.

If you think there could be other features / variables which can add value to your model, then it is a predictive modeling / machine learning problem. For example, if you want to include things like number of Google searches in that category in prior weeks or months or number of people coming to your website (assuming there would be a web presence to your app as well), then you should go for regression models

Hope this helps

Regards,
Kunal


#3

@kunal

Thanks for the tip!

I believe there are indeed other factors influencing the number of users. However, before getting my hands dirty with feature engineering and extraction, I guess I would like to keep it simple initially and build from there.

Thing is, I haven’t really worked with time series before so again I am not sure what are the best methods to employ (e.g. I usually hear things like ‘use ARIMA models’ and stuff like that, but is it really appropriate to use that here?).


#4

Hi kunal,

Can please share your views about Upgrad Big data program with BITS Pilani