Bachelor and Master Theses

Title: Case-based approach for process modeling
Subject: Computer Science
Level: Advanced
Description: Case-based reasoning is a technique to solve new problems based on previous successful cases. Similar problems have similar solutions is the main assumption when case-based reasoning is used. For the problem that the prediction accuracy of real-valued attribute data is not high and some algorithms can be time-consuming, a novel case-based approach is proposed in this article, which combines K nearest neighbor algorithm and regression. This case-based approach is studied with the purpose of improving the efficiency of case retrieval. It was known to be computationally expensive due to the matrix calculation when regression is applied as well as the use of mutual Euclidean distance between elements as a similarity measure. Here, the authors propose to alleviate this drawback by conducting off-line calculation using K nearest neighbor regression method for the existed cases in the dataset. In this way, each case in the dataset can have an extra useful attribute. In the experiment, this case-based approach is applied to a furnace process. After making full use of some updated cases, on-line calculation for the unsolved problem proves to be much faster which owes to the significantly calculation reduction. Some other common methods are also studied in this article. Ten-fold cross validation is applied to show that the prediction performance of this CBR system stays almost the same with the other methods, while the size of the useful updated cases decreases significantly.
Prel. end date: 2013-06-02
Presentation date: 2013-06-13
Student: Bijun Wan
Student: Qin Cao
IDT supervisor: Ning Xiong, +46-21-151716
Examinator: Peter Funk
Peter Funk, +46-21-103153

Rapport och bilagor


Senaste uppdatering



2013-06-12, 00:06

  • Mälardalen University |
  • Box 883 |
  • 721 23 Västerås/Eskilstuna |
  • 021-101300, 016-153600 |
  • webmaster |
  • Latest update: 2018.08.31