What is OCCAM?
OCCAM is software that implements Reconstructibility Analysis. OCCAM combines data mining, machine learning, and statistical analysis in a unique way which lets a user quickly understand the structure of their data, evaluate many possible models, and compare those models under statistical and information-theoretic criteria. Once a model is selected, OCCAM will fit that model to the data, identifying a simplified representation of the data structure that still preserves its essential features, and allow the user to explore and understand variable relationships and make predictions.
How does it work?
A typical OCCAM workflow might look like this:
- Load some data, specifying search options such as the reference model for comparison, types of models to consider, threshold for statistical significance, and depth of search in the lattice of structures
- Use the search results to quickly compare a list of possible models of the data by their entropy or uncertainty, degrees of freedom, statistical significance, AIC and BIC (information-theoretic criteria which compare models by incorporating both the difference in information content and complexity), prediction accuracy, and other measures. (Example search output)
- Once a model is selected, fit that model to the data, and examine it in detail: dependent and independent variable relationships, possible variable states (and the model’s prediction accuracy for each state), prediction rules for each state, statistical significance, and more. (Example fit output)
- Put this insight to use in making predictions, simplifying complex problems while preserving essential features of the data, and answering your important analytical and research questions.
Who can use it?
Data scientists; genomics and health data researchers; economists, sociologists, and other social scientists; systems and complexity science researchers; and in general anyone who is working with categorical data. See the RA and OCCAM research page at Portland State University for more on applications and past and present research using OCCAM.