Back to search

University of British Columbia

Classification capabilities of neural networks : a comparative study using student academic performance

Abstract

dc:description

Among the emerging information technologies, neural networks have been increasingly recognized as a powerful method for classifying and predicting complex data. There have been a number of neural network paradigms being developed. Each paradigm has its own specific features that are applicable to particular tasks. The most popular neural network paradigm among users in the management area is the backpropagation. This paradigm has been extensively tested and proven to outperform traditional techniques on several classification tasks. However, there have been only a few studies determining the comparative capabilities of the backpropagation paradigm to other paradigms that are potentially applicable to the same task. The main purpose of this thesis research is to investigate capabilities and performance of two neural network paradigms: backpropagation and learning vector quantization (LVQ). The other purpose is to prove that neural networks outperform a traditional technique of ordered probit model, which is used as a performance benchmark. In this study, the two neural network paradigms and the ordered probit model are utilized to classify and predict the academic performance of UBC Commerce students. For each paradigm, a number of neural network models with distinct configurations are developed. The first investigation determines how well each paradigm performs in classifying and predicting academic success. The results from running those models on both training and validation samples show that the backpropagation paradigm significantly performs better than the LVQ paradigm in most instances. The second investigation compares the best performance of those paradigms with the performance of ordered probit model. After utilizing the ANOVA to test the statistical significance of difference in prediction performance, the findings show that both backpropagation and LVQ paradigms have higher performance levels than ordered probit model. However, the difference between the performance of backpropagtion and ordered probit model is significant at only the 90% confidence level. On the other hand, the difference between performances of LVQ and ordered probit model is significant at the much higher level of 95%. Essentially, the study has shown that the backpropagation paradigm, on the average, still outperforms the LVQ paradigm in classifying and predicting complex data. The study has also proven that both backpropagation and LVQ are significantly better prediction techniques than the ordered probit approach.

Degree

thesis:*
Name thesis:degree_name
Master of Science - MSc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Business Administration - Management Information Systems
Grantor dc:publisher
University of British Columbia
Year dc:date
1999

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Prompibalcheep, Sansern Art

Rights

dc:rights
Statement dc:rights
  • For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/9188

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Prompibalcheep, Sansern Art. Classification capabilities of neural networks : a comparative study using student academic performance. master's thesis, University of British Columbia, 1999. http://hdl.handle.net/2429/9188