Aplikasi Kernel Dimensional Reduction Untuk Klasifikasi Laporan Penelitian  

Oleh:

Widodo, Agus ; Wasito, Ito

Jenis: Article from Proceeding
Dalam koleksi: Prosiding Seminar Nasional Riset & Teknologi Terapan (Ritektra) "Teknologi Terapan dalam Upaya Meningkatkan Produktivitas dan Daya Saing Industri Nasional", Jakarta 16 - 17 Juni 2010 : Fakultas Teknik Elektro (2010), page 196-205.
Topik: Kernel Dimensional Reduction; Text Mining; Classification; Research Reports
Fulltext: TE-B-09 (Agus W - BPPT Jakarta).pdf (435.34KB)

Isi artikel

Kernel Dimensional Reduction (KDR) is a method of dimensionality reduction for supervised learning developed by Fukumizu et. al. This method will compute a subset of variables from the original set of variables. This dimensionality reduction may provide a simplified explanation and visualization for a human, suppress noise to make a better prediction or decision, and hence reduce the computational burden. Meanwhile, information about research in various fields is now largely available due to the advances of the internet. These research reports usually consist of title, abstract, and several other attributes. The features of the reports can be represented by words or phrase. Hence, the number of features can be quite large. Classifying documents having a lot of features would surely slow down the computation. This study implements the Kernel Dimensional Reduction (KDR) to reduce the number of dimensions of research reports and performs classification on those reduced dimensions. The experiment shows that for two-class classification, the performance of the reduced dimensions by KDR is always better than that of the original dimensions. However, for three-class classification, when the method parameter is set to quadratic programming, the KDR does not outperform the SVM.

Opini Anda

Klik untuk menuliskan opini Anda tentang koleksi ini!

Kembali