Feature Engineering: An Apprentice’s Guide to the “Dark Art” of Machine Learning
summary
Uses IBM bank loan data to improve a RandomForestClassifier (n_estimators=15) by converting categories and extracting dates, raising accuracy from ~68–69% to about an 18% relative gain. Steps include data cleaning, gender→0/1, one-hot encoding of education, weekend flag from effective_date, z‑score standardization of principal, terms and age, and dropping junk columns. Visualization with Seaborn and Yellowbrick informs feature choice. Mentions pandas, scikit‑learn, tsfresh, featuretools and autofeat as automation options.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:04Sorcerers...I mean mentors...
- 1:35What is Feature Engineering?
- 2:56Feature Engineering: Art or Science?
- 4:02Why do we need feature engineering?
- 6:56Three Types of Machine Learning Models
- 8:24First look
- 9:39Our Machine Learning Model: Random Forests
- 10:56Setting up our data for Random Forests
- 11:46Indicate Training Features & Training of model
- 13:10Need to convert Gender to a binary
- 14:01Setting up the test dataset
- 15:53Data Exploration
- 17:25Data Visualization with the Seaborn Library
- 19:311. Adde iustus rudis pars - Adjusting raw features
- 25:28Final Touches - Normalization of Continuous Features
- 26:31The Moment of Truth....
- 26:50Was it worth it?
- 27:28Feature Importances with Yellowbrick
- 28:17Postscript: Python Automated Feature Engineering Tools
related talks
39mFeature Engineering: Unlocking Potential for Predictive Models
01h 28mFeature Engineering Made Simple
02h 17mFeature Engineering is for Everyone!
04h 03mTutorial - Intro to Data Science
43mFeature Engineering for Time Series Forecasting
01h 25mUnlocking the Predictive Power of Relational Data with Automated Feature Engineering
42mCreating correct and capable classifiers
37mCreating correct and capable classifiers