Research Areas ǀ Decision Support and Artificial Intelligence

In this area, we are addressing two fields; decision support and explainable artificial intelligence.

In the field of decision support, our long-term goal is to develop methods and techniques of decision modeling, support them with software and integrate them with data mining systems. We are developing a new generation of DEX software for the construction and use of decision models according to the qualitative multi-criteria method DEX. We are extending DEX with novel leatures (such as integration of qualitative and quantitative elements, advanced analyses of alternatives), implementing it using the state-of-the-art technological design and supporting it in different environments. The software called DEXi Suite currently consists of five components: DEXiLibrary, a central class library implementing DEX, DEXiWin, an interactive desktop program implementing DEX-modelling functionality, and DEXiEval, DEXiPy and DEXiR, packages for using DEX models from the command line, Python and R environments, respectively. In the context of decision support systems for agriculture, we also developed the DEXiWare tool, which is intended for efficient integration of modules that make up decision support systems based on multicriteria decision models.

In the field of explainable artificial intelligence, we are developing and evaluating methods for variable importance estimation and feature ranking in several contexts. This includes novel methods for feature ranking in multi-class and multi-label classification, based on low-dimensional manifold embeddings of the input and output spaces and the Relief method. Similarly, this includes new approaches for unsupervised feature ranking based on the predictive clustering framework as well as the Relief family of methods, on one hand, and based on attribute networks, on the other hand. Moreover, we have developed and evaluated a method for comparing ordered lists (named fuzzy Jaccard index), which can be applied in a range of contexts, including feature rankings, information retrieval, etc.