ML-Based Network Intrusion Detection
Built an intrusion detection pipeline on 6.3M real network flows from CSE-CIC-IDS2018, a public benchmark of 13 attack types • Caught a model that scored perfectly in testing but missed 52K+ DoS attacks from a day of traffic it had not seen • Retrained it to detect all 13 attack types while limiting false alarms, scoring 0.84 macro F1 (1.0 = perfect) • Ran leakage checks to confirm the model learned attack behavior, not shortcuts, and projected daily alert load to test SOC viability

