Predictive analysis

Overview

Predictive analysis is using historical numbers to make predictions about future numbers. It is finding patterns in historical information to create a model of the data and persist those patterns into the future. (1)

Prediction models are meant to forecast individual drivers. E.g. we’re able to forecast revenue by segment, we backtest revenue

Predictive analysis best practices

108.30.20.10 Data Analysis - Predictive analysis best practices

Model classes

Projecting growth/death

• 113.020.020.50 Survival Analysis - Proportional Hazards PH - how many customers will this company acquire in future periods? (models the leading edge of the cohort triangle)
  • Proportional Hazards is in the 113.020.020 Statistics - Survival Analysis family

Projecting repeat/churn

Based on customers already acquired, how many of them will churn? (repeat purchase) - SBG (Shifted beta geometric) - for subscription - 113.020.050 Statistics - BTYD models (Buy till you die) models BG/BB, BG/NBD, Pareto/NBD - Customer-based valuation for non-contractual firms: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3040422

Projecting spend

gamma gamma (not used as of time I arrived at TS), time series

Other approaches

Other approaches to predictive analytics include - CART - MARS - Neural networks

Resources: 1. https://en.wikipedia.org/wiki/Predictive_analytics 2. Re-organization 20200813: This page originally contained links to sub-pages with data for different types of models used for predictive analysis. Content is still here, but I have moved the pages to the 113 Math section.