Starting Your First Predictive Analytics Project
summary
Loads the CUP98 donation data into KNIME, uses a 95k learn set with a 5.1% response (responders give $15 on average, overall mean $0.79 when nonresponders are zeroed) and mailing cost $0.68, partitions training/test, applies correlation filtering, then fits logistic regression (and a neural net attempt), examines ROC/AUC 0.595, notes thresholding problems with default 0.5 and recommends stratified/equal-size sampling, produces a confusion matrix, and exports the final model as PMML export.
AI-written summary · about
chapters (12)
- 0:00Predictive analytics project intro
- 12:04Donation modeling dataset
- 22:15Setting up KNIME
- 36:58Loading data in KNIME
- 52:00Building model from scratch
- 1:06:00Data imputation and variables
- 1:10:33Sampling for training/testing
- 1:26:36Removing correlated variables
- 1:36:12Model performance with ROC
- 1:43:23Logistic regression vs neural networks
- 1:53:17Model predictions and confusion matrix
- 2:00:47Summary and next steps
related talks
56mPredictive Analytics for Practitioners
43mDeveloper Accessible Machine Learning
27mUsing C# for Data Science and Machine Learning
22mPredict the future (or don't) with machine learning in applications
04h 03mTutorial - Intro to Data Science
02h 03mFull stack AI design and development
01h 39mApplying Rigorous Machine Learning Methods in Business Strategy
52mSQL Server Predictive Analytics: Customer Stories