Linear Forecasting of Next Quarter Revenue
In this case, we train the model to predict a single value based on the input time value.
The input will be a simple number describing the time numerically.
We wish to teach it a simple mapping. So, we would like it to be able to map the input value of the so-called 'time-index' to the predicted revenue for any time period.
As a thought experiment, imagine that our revenue increased by the same amount every quarter. For each quarter (time-period/time-step), we could just add some multiple of the timestep value.
Just like the most basic case of y=mx+b where the gradient (m) is how much we increase every period, and x is the time-step number starting at zero.
It's a bit like this with linear regression, so we could input any time-index value like 10, 20 or even 30 which would ideally map forward that many time periods, however it would be inaccurate for such scales.
