Jul. 12 2010 1:30PM

Jul. 12 2010 3:00PM

Thesis supervisors: Jose Antonio Lozano, Inaki Inza and Xabier Irigoien. The application of Probabilistic Graphical Models (PGMs) to fish recruitment forecasting for fisheries management is presented. In particular, it is addressed their combination with supervised data pre-processing methods (missing data imputation, discretization, feature subset selection) and robust error estimation. PGMs have been applied to estimate recruitment scenarios probabilities (low, medium, high). It has been used both, the traditional uni-dimensional approach (one class variable) for single species forecasting and the multi-dimensional approach (multiple class variables) for multiple species recruitment forecasting, where supervised pre-processing methods have been proposed. Naive Bayes for regression approach using kernels density estimation is also used for continuous forecasting to integrate with other quantitative methods. More information at: http://www.sc.ehu.es/ccwbayes/members/jafernandes/

Pravidelný seminář Inteligentní systémy (Contact person: Tomáš Kroupa)