0
$\begingroup$

I am working on project where I want to predict how much customer use my services which translates to Dollar amount. I have data which is having 500 Customers whose monthly usage of services need to predicted for next 12 months.This depends on various factors like current committed usage , Annual Contract Value , happy about our services so on so forth. My question is If I model this as Regression how can I predict for future months like next 20 months , What I have data is till last month.I cannot create future data as most of them more or less remain same. If I want to model this as Time Series dropping all other factors and just looking at Dollar Amount modelling it as Uni-variate I am risking of losing important information. If I want to do it as Multi Variate VAR I am having difficulty featuring in categorical features like SLA Met or not. If I just look into other values like Annual Contract , Committed Contract amount and my target dollar amount then those feature doesn't make sense how can I approach in this case. Below is sample data for better understanding


| Customer | Contract_Annual | Month_Commit | Discount | Month   | True Usage | Open Bugs | Overall Sentiment |
|----------|-----------------|--------------|----------|---------|------------|-----------|-------------------|
| abc_corp | 100000          | 10000        | NO       | JUN2018 | 12000      | YES       | Positive          |
| abc_corp | 100000          | 10000        | NO       | JUL2019 | 15000      | NO        | Positive          |
| :        |                 |              |          |         |            |           |                   |
| abc_corp | 100000          | 10000        | NO       | JUN2020 | 16333      | Yes       | Positive          |
| xyz corp | 250000          | 20833        | YES      | JUN2020 | 100        | YES       | Negative          |
| mnc_corp | 120000          | 10000        | NO       | FEB2020 | 18000      | NO        | Positive          |
| mnc_corp | 120000          | 10000        | NO       | MAR2020 | 18800      | NO        | POSITIVE          |
| :        |                 |              |          |         |            |           |                   |
| mnc_corp | 12000           | 10000        | NO       | JUN2020 | 19000      | NO        | POSITIVE          |

My Question here is If I want to forecast True Usage that is dollar revenue into future 10 months

A) Using Time Series Approach Multi Step Uni-variate assuming current usage reflects everything sentiment into consideration meaning a Customer is spending something on your product based on what your product offers plus his experience which will result in multi step time series month true Usage jun X amount jul Y amount so on and so forth and predict for multiple months

Approach 2) If I want to model this as Multi Variate I am completely confused to take Sentiments into model(as far as I know I cannot account for these ) Month commit , Annual contract more or less remain same with time period it is not multi variate in true sense Approach 3) Some sort of regression where I will make these categorical variables as some sort of encoding and make model aware of these things and also make model aware of contract Annual , Monthly Commit etc but how can I use regression to predict for next 12 months of usage(working example in python could help me to try)

Suggestion on what would be better approach or how these type of use cases can be modeled.I am looking into LSTM as an option or doing it as time series if I could factor in other features , I am hell bent or inclined to try regression and predict for next 15 months but clueless on how I can achieve that

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.