| Abstract : | The ranges of topics that can be studied using statistical methods grow as statisticsscience evolves. A very interesting topic that we will try to analyze statistically andmodeling in this thesis is the material recovery of the aluminum production process asthis done in the ELVAL’s factory. This process is very complex and has many stepswhich affect the quality and the value of the final product. Our data consists ofmeasures which have been made in these steps. Namely, the most important measuresare those which describe the initial weight divided by the produced which is calledReturn Index and the produced weight divided by the initial which called RecoveryIndex. The Return Index will be used in the descriptive analysis of this thesis and theRecovery Index in the model analysis because is located between zero and one. Ourmain goal is to find where the most problems occurred during the production process,which factors affect more the final product and how can we predict the RecoveryIndex better having the minimum error. It is important to emphasize that thedescriptive part of our research will be based on all data whereas modeling will bedone separately for the two types of process, the Hot Rolling and the ContinuousCasting because there are variables which are not defined in both cases. Furthermore,for the descriptive analysis of this thesis which will give us a first view of our dataand some very significant results, will be used simple statistics measures such asPearson Correlation Coefficient, Spearman Correlation Coefficient, Skewness,Kurtosis etc. and figures such as Barplots, Pareto Charts, Box Plots etc. Due to thefact that the diagrammatic representation is a very important part of a statisticalanalysis, the graphs in this thesis will be done with the ggplot library which gives usgreat flexibility. Moreover, the modeling part of this thesis in which we will try tofind which factors affects more the final result and how can we predict it, will beincluded multiple linear models, BIC criterion, penalized regression (Lasso), CARTalgorithms (Regression Tree) and Cross Validation methods. Finally, for theimplementation of this thesis the statistical package which has been used is theprogramming language R. Further analysis and particularities of the problemreferences to chapter Introduction
|
|---|
Copyright © 2013 Library AUEB. All rights reserved.