Case Based Reasoning Group, Research Group, School of Computing, RGU

Case Based Reasoning Group

Case-based reasoning (CBR) is a popular problem-solving methodology because it re-uses the solutions to previously solved problems. But effective CBR systems need good sources of knowledge. Our CBR research concentrates on knowledge acquisition and refinement tools that can assist the different stages of the CBR process; in particular for the retrieval, reuse and revise stages. Automated acquisition and maintenance of case knowledge targets the all important case-base, the main knowledge source of any CBR system.

Projects

CBR for Anomaly Report Processing
Nirmalie Wiratunga, Stewart Massie, Susan Craw

This is a joint project with the European Space Agency (ESA) in Darmstadt, Germany. The project involves the application of CBR to support ESA’s Anomaly Report Processing task. We have currently developed a prototype, CAM, for the case-retrieval stage. CAM maps reports to cases and identifies similar anomaly cases given a new anomaly. The retrieved set is displayed using several alternative report decomposition views. A parallel co-ordinate plot and a spring-embedder assists case comparison highlighting similarities and differences between retrieved reports.

Medical Decision Support with CBR (MedicACE)
Nirmalie Wiratunga, Stewart Massie, Susan Craw

The increasing use of Electronic Patient Records in hospitals provides vast stores of knowledge. MedicACE is a decision support software tool that aids medical diagnosis by applying a Case-Based Reasoning approach using Electronic Patient Records. Records of previous patients who have similar symptoms and test results to the current patient are retrieved and used to suggest a diagnosis. The ability of medical systems to explain their decisions is particularly important. MedicACE incorporates novel visualization tools that aim to make knowledge used by the system more transparent to the user in a bid to instil confidence by providing an explanation of the solution.

Complexity Modelling for CBR Systems
Susan Craw, Stewart Massie, Nirmalie Wiratunga

The contents of the case knowledge container is critical to the performance of classification CBR systems. Our research is developing competence models for CBR systems based on complexity. The complexity measure provides a local indicator of uncertainty within the problem space and is useful in informing case discovery. A complexity-guided case discovery algorithm combines the local complexity measure and boundary identification techniques to actively discover cases close to boundaries. In contrast, a complexity-guided redundancy reduction algorithm uses the local complexity measure to actively retain cases close to boundaries. The algorithm offers control over the balance between maintaining competence and reducing case-base size.

CBR for SmartHouse Installations
Susan Craw, Stella Asiimwe, Nirmalie Wiratunga, Bruce Taylor (SSS)

SmartHouse technology comprises devices that help people with disabilities to retain a level of independence within their homes. CBR designs new SmartHouse installations by reusing SmartHouse solutions for people with similar circumstances. SmartHouse problem-solving experiences are stored as textual reports. Our research builds tools that acquires knowledge for SmartHouse cases from the textual reports. Currently we are building a concept hierarchy of SmartHouse problems to structure the cases and to extract the problem features from the reports.

Knowledge Discovery for CBR
Susan Craw, Nirmalie Wiratunga

Automated tools have been developed to acquire retrieval and adaptation knowledge. These tools have been applied to a challenging tablet formulation domain where CBR is used to design recipes for tablets. Additional ingredients are required to balance the physical and chemical properties of the drug to enable the manufacture of the tablet, its handling by the patient, and disintegration after swallowing. Knowledge-light learning uses only the cases in the case-base to learn explicit knowledge to improve retrieval and achieve adaptation. Funded by EPSRC. Collaborators: AstraZeneca, ISoft.

Tablet Formulation Using Cases or Rules
Susan Craw, Robin Boswell, Nirmalie Wiratunga

Funded by EPSRC. Collaborators: IntelliCorp, AstraZeneca, LogicaCMG.