SHAP-Tree-MCDA
SHAP-Tree-MCDA is an explanation module for hierarchical Multi-Criteria Decision Aid (MCDA) models.
SHAP-Tree-MCDA is an explanation module for hierarchical Multi-Criteria Decision Aid (MCDA) models. MCDA aims at representing the preferences of a user on the basis of multiple and confliciting criteria. SHAP-Tree-MCDA uses a tree representation to allow the user to identify the most contributing criteria in a given decision, using an extension of the Shapley value.
Execution environment: windows
Version of the environment: W7 & W10
Install & Run: Unzip directory and run Myriad.exe
More information: Multi-Criteria Decision Aid (MCDA) aims at representing the preferences of a user on the basis of multiple and confliciting criteria. We consider utility models where the preferences is quantified by a utility. The MCDA model is represented as a tree in which the root node is the overall evaluation (utility), the leaves are the elemenatry criteria, ad intermediate nodes are aggregation functions. The interest of these intermediate nodes is to organize the set of criteria into group of coherent concepts, which helps to explain the decision. The explanation module aims at computing the level of contribution of each node in the tree when comparing two options described by a value over each atribute. This allows the user to identify the most contributing criteria in a given decision.
The specific contribution of our asset is an extension of the Shapley value on trees. The Shapley value is a well-known concept in Cooperative Game Theory that fairly allocates the total worth gained by a set of players among themselves. This algorithms implements the following paper:
[1] C. Labreuche and S. Fossier. Explaining multi-criteria decision aiding models with an extended shapley value. In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI 2018), pages 331–339, Stockholm, Sweden, July 2018.