An Intuitive Guide to Relevance-Based Prediction

By Megan Czasonis, Mark Kritzman, and David Turkington

Relevance-based prediction is a new approach to data-driven forecasting that serves as a favorable alternative to both linear regression analysis and machine learning. It follows from two seminal scientific innovations: Prasanta Mahalanobis’ distance measure and Claude Shannon’s information theory. Relevance-based prediction rests on three key tenets:

1) relevance, which measures the importance of an observation to a prediction;

2) fit, which measures the reliability of each individual prediction task;

3) codependence, which holds that the choice of observations and predictive variables should be determined jointly for each individual prediction task