One of many frequent issues in time-series evaluation is lacking information.
As we now have seen in Part 1, easy imputation methods or regression-based fashions like linear regression and choice timber can get us a good distance.
However what if we have to deal with extra refined patterns and seize fine-grained fluctuations in advanced time-series information?
On this article, we are going to discover how a Neural Community (NN) can be utilized to impute lacking values.
The strengths of NNs are their functionality to seize nonlinear patterns and interactions in information. Though NNs are often computationally costly, they’ll supply a really efficient technique to impute lacking time-series information in circumstances the place easier fashions fail.
We are going to work with the identical dataset as in Half 1 and Half 2, with 10% values lacking, launched randomly for the mock vitality manufacturing dataset.
Don’t miss out Half 1 of this collection: