SVMs are outdated for time series classification - Citation needed I'm looking for a strong publication I can cite to prove that SVMs perform worse for time series classification tasks in comparison to other methods (such as ANNs, Decision Trees, Gradient Boosting Machines, etc...).
I found several posts indicating or justifying that SVMs are not popular nowadays, that deep learning rendered kernel methods obsolete or that SVMs have severe drawbacks (such as parameter sensitivity or the need to conduct a grid search for the parameters).
However, I could not find a source explicitly stating that SVMs perform worse for time series classification tasks than more recent models.
 A: Regarding

I'm looking for a strong publication I can cite to prove that SVMs perform worse for time series classification tasks in comparison to other methods (such as ANNs, Decision Trees, Gradient Boosting Machines, etc...),

you might be seeking a reference for a claim that is not generally correct. A key thing to remember is, there is no free lunch in model selection. As Dikran Marsupian says in this answer,

the no free lunch theorems say that there is no a-priori superiority for any classifier system over the others, so the best classifier for a particular task is itself task-dependent

and adds:

However there is more compelling theory for the SVM that suggests it is likely to be better choice than many other approaches for many problems.

I bet there exists a data generating process and a sample size for which SVM systematically beats other methods.

Regarding

I found several posts indicating or justifying that SVMs are not popular nowadays, that deep learning rendered kernel methods obsolete or that SVMs have severe drawbacks (such as parameter sensitivity or the need to conduct a grid search for the parameters),

this might be true as of now. Perhaps a lot of time series that people are forecasting and writing about lend themselves more easily to those other methods. But popularity and optimality do not always go hand in hand, and the no free lunch theorem remains applicable.
A: I know, this might be a late answer, but I just wanted to add that you cannot simply say that SVMs are outdated for time series classification (TSC).
It rather depends on the kernel function you're using. The conventional kernels such as RBF, poly, etc. sure aren't fitting well for TSC. This is due to the time axis distortion problem

Because of the time axis distortion problem, classical
kernel functions, such as Gaussian RBF (GRBF) and
polynomial, generally are not suitable for SVM-based time
series classification. Motivated by the success of dynamic
time wrapping (DTW) distance, it has been suggested to
utilize elastic measure to construct appropriate kernel.

This said, if you choose a custom kernel function such as GTWED or GERP (subtypes of GEMK), SVMs show high potential in TSC:

Experimental results show that, in terms of
classification accuracy, SVM with GEMK is much superior
to the state-of-the-art similarity measure methods and SVM
with GRBF and GDTW kernels.

Source: Dongyu Zhang, Wangmeng Zuo, David Zhang, Hongzhi Zhang: Time Series Classification Using Support Vector Machine with Gaussian Elastic Metric Kernel, 2010. In: 2010 International Conference on Pattern Recognition. DOI: 10.1109/ICPR.2010.16
(https://ieeexplore.ieee.org/abstract/document/5597650)
