0
$\begingroup$

I am working on a 'two-class classification of multivariate time-series' problem. I used two different approaches:

1) Manual time-series feature engineering (such as slope, intercept, variance, etc. using tsfresh python library) followed by a multi-layer perceptron.

2) Conv1D followed by LSTM followed by Dense layers (keras/Python) without any feature engineering.

They both yield very similar results (~40% precision, ~95% recall).

Can I infer anything from the similarity of results from two different architectures?

$\endgroup$
0
$\begingroup$

I believe the only way to make a science-backed inference, would be if you fine-tuned both of your proposed solutions, to ensure that both are performing at their maximum potential. Only then you can say that method A is better than method B, otherwise it might just be a matter of improper selection of parameters that leads to those results.

However, if both methods still yield similar results after you have fine-tuned them, then you can follow Occam's razor and select the simplest one as the best one. By simplest, you can consider the one with the least amount of parameters and least time needed for training/inference.

$\endgroup$

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.