Business Context

NetSecure, a network monitoring company processing millions of connection logs per day, wants a lightweight classifier that labels IPv4 addresses as private or public before downstream traffic analysis. The goal is not to hand-code rules only, but to evaluate whether a production ML pipeline can learn the distinction reliably from parsed IP features and remain easy to deploy.

Dataset

You are given a labeled IPv4 dataset built from historical network logs and enrichment jobs.

• Size: 240K IPv4 records, 18 features
• Target: Binary label — private (1) vs public (0)
• Class balance: 28% private, 72% public
• Missing data: <2% missing in metadata fields; no missing octets after parsing validation

Success Criteria

A good solution should achieve F1 >= 0.98, precision >= 0.99 on the private class, and near-perfect recall on standard RFC1918 private ranges. The model should also expose feature importance or simple decision logic for auditability.

Constraints

• Inference must score at least 50K IPs/sec in batch jobs
• The solution must be interpretable enough for network analysts
• Retraining should be simple and inexpensive
• Avoid leakage from labels encoded directly in engineered features

Deliverables

1. Build a binary classification pipeline to predict whether an IPv4 address is private or public.
2. Explain feature engineering from raw dotted-quad strings to model-ready inputs.
3. Compare at least one interpretable baseline with one tree-based model.
4. Evaluate using class-specific precision, recall, F1, and confusion matrix.
5. Describe how you would deploy and monitor the model in a log-processing pipeline.