Gone Phishing: Lessons Learned Training a Phishing Email Classifier with Highly Imbalanced Biased Labels

Explore the challenges and solutions in training a machine learning classifier for phishing email detection with imbalanced and biased data in this 49-minute conference talk from the Data Science Festival. Delve into the real-world issues faced by Chris Ballard, Lead Data Scientist at Tessian, as he discusses the importance of high-quality labeled training and test data in machine learning with imbalanced datasets. Learn how to overcome selection bias in label generation, handle noisy data, and develop reliable evaluation procedures for assessing model performance before deployment. Gain practical insights on dealing with highly imbalanced data, considering factors beyond precision and recall, and adapting evaluation methods based on label sourcing techniques. Apply these lessons to your own projects and understand the critical considerations when working with imbalanced datasets in real-world machine learning applications.