What is the best ML/DL model to choose to calculate mobile network utilization increase

Question

Let’s imagine the following scenario:

The marketing department decides to do the promotion next month and would like to give to every single mobile customer extra 10GB of data.

Due to simplicity, I have the following 5 features:

A: Amount of data volume downloaded by all customers per base station/transmitter (MB)

B: Number of connected customers to the base station (#)

C: Total duration of the connected customers to the base station (seconds)

D: Radio resource utilization (%)

E: Average throughput per customer (kbps)

Considering the marketing requirements, I can work out that the average increase of data volume per base station is 10% (Feature A).

The question is: What impact is this promotion going to have on the mobile network (every single base station), mainly on Radio resource utilization (Feature D) and the average throughput per customer (Feature E).

I have tons of data for each base station from low to high traffic scenario, but I cannot train the model with 10% extra data volume traffic increase as this situation never happened in the network. It is still going to be "the peak data volume per Base station so far" + extra 10%.

How can I train the model, if target label is unknown (peak data volume + extra 10% never happened on the particular base station and I cannot take the data/stast from a different base station as the traffic pattern is different)?
It would be enough to point me into the direction, I can find more info and to study it further. Thanks.

Answer 1

I do not see why having a percentage greater than 100% might be a problem to you? You might normalize it by dividing by 100, so 100% might become 1. That feature will be your label. I suggest you try several regression algorithms, with different settings, and choose the best model i.e. the model which provides the best result. You have a bunch of regressors in scikit learn, you may also try XGBoostRegressor or CatBoostRegressor.