Where academic tradition
meets the exciting future

You are here: TUCS > PUBLICATIONS > Publication Search > A Two Level Approach to Making...

A Two Level Approach to Making Class Predictions

Adrian Costea, Tomas Eklund, A Two Level Approach to Making Class Predictions. In: Ralph H. jr Sprague (Ed.), Proceedings of the 36th Hawaii International Conference on Systems Sciences (HICSS-36), 1–10, IEEE, 2003.

Abstract:

In this paper we propose a new two-level methodology for assessing
countries’/companies’ economic/financial performance. The methodology is
based on two major techniques of grouping data: cluster analysis and
predictive classification models. First we use cluster analysis in terms
of self-organizing maps to find possible clusters in data in terms of
economic/financial performance. We then interpret the maps and define
outcome values (classes) for each data row. Lastly we build classifiers
using two different predictive models (multinomial logistic regression
and decision trees) and compare the accuracy of these models. Our
findings claim that the results of the two classification techniques are
similar in terms of accuracy rate and class predictions. Furthermore, we
focus our efforts on understanding the decision process corresponding to
the two predictive models. Moreover, we claim that our methodology, if
correctly implemented, extends the applicability of the self-organizing
map for clustering of financial data, and thereby, for financial
analysis.

Files:

Abstract in PDF-format

BibTeX entry:

@INPROCEEDINGS{inpCoEk03a,
  title = {A Two Level Approach to Making Class Predictions},
  booktitle = {Proceedings of the 36th Hawaii International Conference on Systems Sciences (HICSS-36)},
  author = {Costea, Adrian and Eklund, Tomas},
  editor = {Sprague, Ralph H. jr},
  publisher = {IEEE},
  pages = {1–10},
  year = {2003},
}

Belongs to TUCS Research Unit(s): Data Mining and Knowledge Management Laboratory

Publication Forum rating of this publication: level 1

Edit publication