I have a CSV file filled with user activity data. Each row gives me a timestamp, an identifier, a free-text message, and a label that marks whether the line is normal or malicious. What I need is a working intrusion-detection system that relies on entropy calculated over configurable time intervals rather than on signature matching. Your task is to design and code the complete detection pipeline: read the CSV, parse the timestamps into evenly spaced windows, compute Shannon entropy (or an equally sound metric you may justify) for the identifier, message, and any other derivative you propose, then flag the windows that deviate from the learned baseline. I’m comfortable with Python and its usual stack (pandas, NumPy, SciPy, scikit-learn, Matplotlib), so please build on that unless you have a compelling alternative. The final result should let me adjust the window size, the entropy threshold, and the columns included, then produce a clear report of the anomalous intervals along with a confusion-matrix style summary against the provided labels. A short README explaining how to run everything and how each parameter influences detection rounds things off. Acceptance criteria • Script runs from the command line, taking my CSV path as an argument. • Generates a CSV or JSON report listing suspicious time windows and their scores. • Reaches an F1 score that beats a naïve random detector by a meaningful margin (show the comparison). • All code, plots, and documentation are delivered in a clean Git repository.