This session is dedicated to an introduction of (artificial) neural networks and discusses a basic network architecture for classification, the (multilayer) Feedforward Neural Network (FNN), and an unsupervised network, the AutoEncoder (AE), which can be used in a classification setting.
For feature importance, you can use tools we used in other BIDS tutorials, or another package called shap, install via:
conda install -c conda-forge shap # alternatively do pip install shap
Have a play around with some toy data and neural networks in a browser: playground