Shadow-A2234/Random-Forest-WAF ? reverse-engineered prompt

Build me a simple web application that can learn to spot suspicious or malicious request payloads from a CSV file and then use that model to check new requests.

I want a small Python project with a training script, an evaluation script, and a basic server that loads the saved model and makes predictions. The training should read the payload data, build the model, save it, and the evaluation should show how well it performs with clear results like a confusion matrix and feature importance. The server should be easy to run locally and return whether a request looks safe or risky.

Please keep it straightforward and practical, with sensible defaults, so I can retrain the model and test it without a lot of setup. If you need anything unclear from the data format or model behavior, make a reasonable assumption and keep going.