Decision Tree Classification od Products Using C5.0 and Prediction of Workload Using Time Series Analysis

Abstract: This thesis covers the analysis of manually classified information, with afocus on classification using decision trees and prediction using time seriesanalysis. In the classification part of the thesis, an existing manual classificationis evaluated and compared to the classification obtained with adecision tree approach.In this thesis, the classes are comparable to each other, i.e. each class can beassigned a numerical value. Thus, the manual classification can be comparedto the decision tree classification with respect to the distance from the trueclass. The results show that decision tree classifications tend to fall intoneighboring classes with some exceptions.Using time series analysis, the daily rate of items arriving to a repair workshopis evaluated and predicted. The result shows that it is possible to finda predictor for the arrival rate of workload. This is performed by implementinga classical decomposition model to forecast a general trend and seasonalchanges, and improving the predictions by fitting a linear dynamical modeldriven by a white noise process. An automated algorithm to update thismodel is implemented to minimize the maintenance of forecasting.