If you want to organise lots of data in an understandable format (and predict the future besides), you can’t beat decision trees. (Except with Bayesian networks, artificial neural networks, case-based reasoning, or transformation-based learning.) You can throw tons of data at a decision tree algorithm, and it’ll present it as an easy-to-read chart.
Here’s an interesting decision tree about Hillary v Obama in Pennsylvania. The most discriminating features (ones that provide the most information) appear at the top of the tree. The way this one turned out, it appears that Obama wins in counties that are better educated and/or have a higher concentration of black voters.
Recent Comments