Section: Research Program

Decision support

Patterns, especially predictive sequential patterns, resulting from mining a dataset have often a direct application in diagnosis. Lacodam inherits from the former Dream team a strong background in decision support systems, with an internationally recognized expertise in diagnosis. This AI subfield is concerned with determining if a system is operating normally or not, and if the system is in an abnormal state, to determine the cause of the faulty behavior. The considered system can as well be an agro- or eco-system, a software system or an animal or human being, as well.

The increasing volumes of data coming from a wide range of different systems (ex: sensor data from agro-environmental systems, log data from software systems, biological data coming from health monitoring systems) show that it is possible to gather more and more observations for such systems. Thus, it should be possible to exploit such observations to help human or software agents to take better decisions. Hence, while keeping the strong interest on decision support (and especially diagnosis) that existed in Dream, Lacodam adds the idea that the decision support systems should take advantage of the huge volumes of data available. This third and last research axis is thus a meeting point for all members of the team, as it requires to integrate AI techniques of traditional decision support systems with results from data mining techniques.

Two main research axes are investigated in Lacodam:

• Diagnosis-based approaches. We are exploring how to integrate knowledge found from pattern mining approaches, possibly with the help of interactive methods, into the qualitative models. The goal of such work is to partly automate the construction of the model, which can require a lot of human effort otherwise.

• Actionable patterns and rules. In many settings of “exploratory data mining”, the actual interest of a pattern is hard to assess, as it may be hard to measure or may be subjective (resulting from introducing the user in the mining process). However, there exist applications where once patterns are found, there are well defined measures to define what this pattern will bring to the user. Further, patterns and rules that can lead to actual actions beneficial to the user are called actionable patterns. Such actionable patterns and rules are especially important for industry.