Python implementation of a data perturbation method to determine relevant features for NN learning
Ardelean, E.-R.; Terec, R.-D.; Marieş, C.-M.; Moca, V.V.; Mureşan, R.C.; Dînşoreanu, M. Spike Sorting Using Superlets: Identifying Feature Importance through Perturbation. In Proceedings of the 2023 IEEE 19th International Conference on Intelligent Computer Communication and Processing (ICCP); October 2023; pp. 357–362.
This plot shows the most important features in learning for the classification of neuronal spikes in the form of spectrograms. Higher values represent a bigger difference between the perturbed set and the original set, thus, higher values represent more important features.