Clus-Hyper: a hyper-heuristic method to generate decision tree-induction algorithms tailored to HMC problems

Funding: 
BOF
Period: 
1 October, 2015 to 30 September, 2016
DTAI Participants: 
Partners: 
Celine Vens (KU Leuven-Kortrijk)
Ricardo Cerri (UFSCar Brazil)
Marcio Basgalupp (UNIFESP Brazil)

The overall theme of this project is to improve data mining or machine learning algorithms that construct predictive models for complex targets. In particular, we will focus on hierarchical multi-label classification (HMC) problems, which have many applications in the biomedical domain. For instance, gene function prediction and medical image classification are instantiations of HMC problems: in these tasks we want to annotate an instance (gene or image) with multiple classes (gene functions or image annotations), where these classes belong to a hierarchically structured vocabulary (e.g. using the Gene Ontology for gene functions and the IRMA codes for image annotations). The investigators at KU Leuven and Federal University of São Carlos are leading experts in this domain, each from their own specific background and expertise at the algorithmic or application side.

To reach this goal we intend to use the expertise of the investigator at Federal University of São Paulo, and design a hyper-heuristic algorithm for HMC. More precisely, we will extend an existing hyper-heuristic algorithm for decision tree learning to the HMC context.

Group: