Beta 1


Title Comparing heuristic approaches for the prediction of noisy time series
Author Boss, Niklas Skamriis
Supervisor Witt, Carsten (Algorithms and Logic, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2010
Abstract The typical study of the performance of heuristics for time series predictions, tests only a single approach. This thesis describes the work on three heuristics for time series predictions: Genetic Programming, Neural Networks and Particle Swarm Optimization. The aim is to investigate how the performance of the heuristics degrades as noise in the input data increases, in both theoretical time series and time series of real scenarios. A thorough description of the heuristics is given, to give an insight in the important mechanisms of the heuristics. This sets the ground for the choices during the experiments, and provides the knowledge to support the results and observations from the experiments. The objective of the practical work is to implement a framework for creating multiple investigations on how random noise affects the predictive capabilities of the heuristics. The noise will be added to the input data to increase the difficulty of the prediction tasks. The quality of the predictions will be compared across three time series using the Mean Square Error. The results will form the base for a conclusion of some general tendencies, as well as recommendations for future work.
Imprint Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Series IMM-M.Sc.-2010-30
Fulltext
Original PDF ep10_30.pdf (2.29 MB)
Admin Creation date: 2011-01-19    Update date: 2011-01-19    Source: dtu    ID: 274262    Original MXD