See https://www.kaggle.com/kdubey/sparse-linear-svm-is-accurate-and-interpretable for a full write-up and code.
There, I do my best to throughly analyze a model's stability and use, and communicate shortcomings of modeling and simulation strategies.
This repo simply houses a script that can be run to easily reproduce all results.