- 021 808 3244Location:
Department Statistics & Actuarial Science, Van der Sterr Building, Room 2048
The study presented looked at possible methods and processes involved in the imputation of complete missing blocks of data. A secondary aim of the study was to investigate the accuracy of various predictive models constructed on the blocks of imputed data. An iterative bagging technique applied to variants of the neural network, decision tree and multiple linear regression (MLR) improved the estimates produced by the modelling procedures. A stochastic gradient boosted decision tree (SGBT) was also constructed as a comparison to the bagged decision tree.
The results indicated that the choice of an imputation method as well as the selection of a predictive model is dependent on the data and hence should be a data-driven process.