Speaker: Omer Sagi, Ben-gurion university of the Negev
Growing a tree from a forest: Transforming a decision forest into an explainable tree
Ensemble methods considered today as the state-of-the art in solving plethora of machine learning challenges. A particular type of popular ensemble method is Decision Forest which is an ensemble of decision trees. However, despite their high degree of accuracy, other models may be preferred over ensemble models as it is impossible to interpret ensemble predictions and difficult to acquire knowledge from the ensemble inner structure. such issues of explainability deter users from using ensemble models in domains that require a clear and rational explanation for individual decisions (e.g. medicine, insurance, etc.) and when an explanation for unusual classifications is required in order to make a decision. This work presents a set of methods for making decision forests more explainable while its main contribution is in presenting a method for generating a decision tree from a pre-trained decision forest. We call the new developed model a Forest-Based Tree (FBT). It is generated by applying an algorithm that uses the inner structure of a decision forest towards creating a new intelligible model in the form of a single decision tree, a model that considered intelligible and suitable for tasks that require explainable results.
Omer Sagi is a Ph.D. student in the Software and Information Systems Engineering department at Ben Gurion University, his research addresses the subject of explainability of machine-learning models, focusing on making ensemble models more explainable. Omer received his B.Sc. and M.sc. degrees from the department of industrial engineering and management of Ben-Gurion university. In addition to his research Omer works as a lead data scientist at Honeybook, a software as a service solution for small businesses at the USA. Prior to HoneyBook, Omer was working for 5 years as a senior data scientist at Dell technologies.