Probabilistic Forecasting of Binary Events using Regression
Forecasting the probability of extreme values with the cumulative distribution function
In this article, we’ll explore the probabilistic forecasting of binary events in time series. The goal is to predict the probability that the time series will exceed a critical threshold.
You’ll learn how (and why) to use a regression model to compute binary probabilities.
Introduction
First of all, why would you use regression to compute binary probabilities instead of a classifier?
The probabilistic forecasting of binary events is usually framed as a classification problem. But, a regression approach may be preferable for two reasons:
- Interest in both the point forecasts and event probabilities;
- Varying exceedance thresholds.
Interest in both the point forecasts and event probabilities
Sometimes you may want to forecast the value of future observations as well as the probability of a related event.
For example, in the case of forecasting the height of ocean waves. Ocean waves are a promising source of clean energy…