Thanks for pointing out that Linear and Logistic regression are under-rated, understandable and can help avoid the pitfalls associated with coefficient explosion. A big plus, given that there are lots of people in Customer Success teams and Business Analytics who don’t “like to code”, is that you can actually do a lot of exploration and obtain insights using Excel and GRG non-Linear.
The analogy I make with neural networks with many layers and much depth is to the Ptolemaic model — more than 8 epicycles per planet and lots of calculational techniques and tons of data. Of course it had great predictive power, and if it didn’t predict accurately enough one just added another epicycle! But this completely obscured the fact that you really only needed only 1 + 3 per planet coefficients since the Ptolemaic coefficients had very high covariances (in modern terms).