Please join us for a seminar talk with Giles Hooker.

Decision Trees and CLT’s: Machine Learning and Ecological Inference

Abstract

Machine learning methods are powerful statistical tools that produce prediction models from data. These methods have enjoyed a great deal of success over the past few decades in a wide range of application areas. Within ecology, they have been most widely used in species distribution models such as MaxEnt.

However, while they are very effective at prediction, the models produced by machine learning algorithms tend to act as black boxes: they result in algebraically complex expressions that are not accessible to human interpretation. They also do not come with the familiar quantification of uncertainty in terms of confidence intervals or hypothesis tests.

In this talk, I will use citizen science data collected though Cornell’s Laboratory of Ornithology Ebird program to demonstrate new developments in methods that allow random forests – one of the most successful machine learning method – to provide ecological insight that can be paired with formal statistical inference. These developments point to new ways to use large and complex data to better understand broad scale ecology.

‍