Predictive Analytics for Practitioners
summary
Practical roadmap for building deployed predictive models: emphasis on data preparation (49 raw variables + 18 demographics + 79 derived → 93 columns, 21,557 rows), sampling into train/validation/test to avoid overfitting, and comparing algorithms (nearest neighbor, neural nets, naive Bayes, logistic regression, decision trees). Uses TIBCO Spotfire Miner (formerly Inle Miner) for cleaning and feature engineering, scores 2 million prospects and hits 13.67% response where 13.5% was required, using ROC/lift to choose the cut.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:25Technical Assistance
- 3:21What do Predictive Modelers do? The CRISP-DM Process Model
- 7:37CRISP-DM: Business Understanding Steps
- 10:37Objective's
- 12:03Data Understanding Steps
- 14:00Source Data
- 15:25Data Preparation (Conditioning) Steps
- 16:39Data Size
- 18:04Modeling Steps
- 19:07Sampling
- 24:18Classifiers Find Different Decision Boundaries
- 27:43Assess Models: ROC Curves
- 31:09CRISP-DM Step 6: Deployment Steps
- 31:35Model Results after Deployment
- 34:41What is Predictive Analytics? Simple Definitions
- 40:25Predictive Analytics vs. Data Science
- 43:52What Degree Does it Take to Be a Predictive Modeler?
- 47:08The Analyst's Journey
related talks
52mSQL Server Predictive Analytics: Customer Stories
22mPredict the future (or don't) with machine learning in applications
01h 39mApplying Rigorous Machine Learning Methods in Business Strategy
55mWhat Skills Do Predictive Modelers Need?
02h 03mStarting Your First Predictive Analytics Project
02h 00mData Science Driven Business Insights with Python and PySp
53mMachine Learning in E‑commerce
10mIs Bigger Really Better? Predictive Analytics with Fine-grained Behavior Data