What does rmse of a LSTM model tells?

Question

What does rmse of a LSTM model tells?

Stupid_Intern

2022年2月15日 12:21

Suppose I made a model which has rmse of 50 Now when I predict the next data which is 500

So does that mean the actual value has high probability to be within the range of 450 - 550 ? If so what is the probability that it will be in this range?

Or it means the actual value has high probability to be within the range of 475- 525 ? If so what is the probability that it will be in this range?

Topic rmse lstm time-series

Category Data Science

Fnguyen · Accepted Answer · 2022年2月15日 12:21

RMSE is calculate based on multiple predictions so it isn't as easy as saying when you predict 500 the actual result will be a deterministic calculation of 500+-RMSE.

Instead you would expect that with a new batch of e.g. n = 500 predictions, the square-root of the mean squared difference would be close to 50.

Note that due to the root and squaring operations, a RMSE indicates an absolute average difference around 50 which can be interpreted as 500 +-50 not 500+-25.

In practice individual results of course can vary much more e.g. for:

y_pred  y_act
50      100
50      100
50      100

the RMSE is the same as

y_pred  y_act
50      50
50      50
50      136

and so on. This means that you can use normal distribution assumptions to say:

"Assuming my errors are not biased and normally distributed, my range of error should be +-50 for 68% of my values."

What does rmse of a LSTM model tells?

About