Week 5

In previous case, we only use 1 lag to model AR ,MA and ARMA models

Below graph shows the variation of means square error for the test set, by changing position of lag which is used

So far we use only one lag, however, water level prediction can depend on more than one lag. In order to evaluate that hypothesis, a new Linear auto regressive model was created using 100 lag window

Performance of above model as follows

Problem encounters with this model

• We can still only predict one time step ahead

• Variance of errors is too large.

• At peaks error is comparatively large