|Institutional Office Address
DMIF, Room NS3
|Institutional Laboratory Address
Data science lab
|Research Project Title
Data Mining with temporal aspects
|Research Project Description
Data mining is a set of tools and techniques aimed at extracting meaningful patterns or regularities from data. Such patterns are captures by models, which are trained by means of machine learning algorithms.
Learning tasks can be distinguished into two main categories: unsupervised and supervised. According to the former, data is explored for general regularities, as in the case of clustering or association rules learning. The latter focuses on the discovery of a relationship between a set of predictors and a dependent variable, as in the case of linear regression or decision tree induction. Such a relationship can then be used to predict the dependent variable for cases in which it is unknown.
In both areas, temporal data mining is a rapidly-developing field or research, since in many domains sequential and correlated events can be observed.
Nevertheless, there is a lack of study on methods capable of exploiting in a human-readable fashion both temporal and atemporal data, for prediction purposes. For example, decision trees are one of the most used models for supervised tasks, owing their popularity mainly to their interpretability and efficiency. However, they currently lack the ability of dealing with mixed temporal and atemporal data.
A possible research direction is thus a temporal extension of decision trees. On the basis of such an extended model, it is then possible to conduct more advanced tasks, such as temporal feature selection via wrapper strategy.