I can give you some pointers on the first that will help you in your quest. Let me get back to you later this week.

Here is a document that walks through the scenarios, and shows comparative differences. I've also included a bonus item on interpretating PCA components with IBM Bob.  Good luck!

Attachment(s)

PCA_TSNE_Comparison.docx   1.18 MB 1 version

The trade-offs are debatable. however, if the goal is a downstream classification model where interpretability, reproducibility, dimensionality reduction and stability matters, i would prefer PCA. PCA is mathematically or statistically well grounded. it allows us to quantify how much variance each component explains and understand variable contributions. The resulting components are uncorrelated which can help address multicollinearity and reduce dimensionality, noise and computational costs. PCA is deterministic and provides a consistent transformation making it stable in production pipelines than t-SNE. 

t-SNE is mainly exploratory/visualization technique.

Great comparison! PCA is usually the better choice when you need interpretability, lower computational cost, and stable features for downstream tasks like classification. t-SNE is more useful for visualizing complex nonlinear structures and local clusters, but its embeddings can be harder to interpret and aren't generally ideal as features for a classifier. For creative visualization ideas alongside data-focused projects, you can also explore trending CapCut templates on Editing Prompt.