Detecting Deviating Cells

This thesis is a replication and extension to the paper: Detecting Deviating Cells (DDC) by Rousseeuw and Van den Bossche (2016). DDC is a new cellwise outlier detection method, that outperforms all other known outlier detection methods. In addition, the method can replace outliers in a data set by estimated values. The DDC algorithm is analyzed and implemented in the software program MATLAB, in order to replicate and validate the results and implementation of the authors. In addition, the DDC method effects on multi-linear regressions are evaluated. The preliminary conclusion is that the DDC method improves the prediction power of linear regressions, but falls short in comparison to robust regressions.