Title: A methodology for mining material properties with unsupervised learning

Authors: Hisham Al-Mubaid, Emad S. Abouel Nasr, Mohammed Hussein

Addresses: Department of Computer Science, University of Houston – Clear Lake, 2700 Bay Area Blvd., Houston, TX, 77062, USA. ' Mechanical Engineering Department, Helwan University, Helwan, Cairo, Egypt. ' Mechanical Engineering Department, Helwan University, Helwan, Cairo, Egypt

Abstract: Studying material properties from a data mining perspective can be beneficial for manufacturing and other industrial engineering applications. We present new ways to utilise data mining and machine learning in analysing material properties with experimental study. This work employs an effective feature reduction technique with clustering and classification to extract the most significant properties of the materials which can benefit various manufacturing industries. We conducted the experiments on five material datasets. The evaluation results proved that: 1) The feature reduction technique is quite effective in reducing the features to 20 or ten by removing all non-significant and redundant features. Thus, this model represents an effective way of extracting the most significant features for each class of materials. 2) The material databases and properties can be easily and feasibly analysed and examined in data mining and the outcomes can be very useful for the manufacturing industries and other industrial engineering applications.

Keywords: material properties; data mining; unsupervised learning; manufacturing; industrial engineering; machine learning; feature reduction; clustering; classification.

DOI: 10.1504/IJRAPIDM.2009.029385

International Journal of Rapid Manufacturing, 2009 Vol.1 No.2, pp.237 - 252

Available online: 28 Nov 2009 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article