UUM Electronic Theses and Dissertation
UUM ETD | Universiti Utara Malaysian Electronic Theses and Dissertation
FAQs | Feedback | Search Tips | Sitemap

Predicting Diseases Using Multi-BackPropagation

Wan Hussain, Wan Ishak (2002) Predicting Diseases Using Multi-BackPropagation. Masters thesis, Universiti Utara Malaysia.

[thumbnail of WAN_HUSSAIN_B._WAN_ISHAK.pdf] PDF
WAN_HUSSAIN_B._WAN_ISHAK.pdf
Restricted to Registered users only

Download (13MB) | Request a copy
[thumbnail of 1.WAN_HUSSAIN_B._WAN_ISHAK.pdf]
Preview
PDF
1.WAN_HUSSAIN_B._WAN_ISHAK.pdf

Download (1MB) | Preview

Abstract

A Computer-based medical system plays an important role in the current practice of medicine. Initially, computer is used to store and manage information effectively.
The computer becomes more important with the introduction of the intelligent system. The intelligent medical system increases the ability of medical practitioners
in providing diagnosis and prognosis. Neural network is one of the artificial intelligence techniques that emulate the human neuron function. Neural network enable the computer to "learn" and "think" like human. However, learning usually involves a large amount of data. If more data is used, the network complexity will be increased. Complex network is hard to learn and take more time to generalize. Thus this study proposed a multi-network approach as oppose to the single network approach. Multi-network approach does not require any changes in neural network learning algorithm. Instead, the large data is divided into several smaller categories or network. Both approaches are tested and compared. The results show that the estimation time for the single network with 26 variables based on 7466 data set is approximately 1,037,472,836 milliseconds to complete the learning with 100 percent generalization performance. On the other hand, based on 256 data sets the network takes 2,459,172,864 milliseconds to complete the learning. The epochs are estimated as 359,544 and 26,2 14,400 respectively.
In the multi-network approach, five different networks and one integration network were constructed. The experiments showed that all six networks managed to learn the data completely in only several epochs. The time taken by the networks are 281, 197, 32, 440, 83 and 22 respectively for the risk factor, medication, investigation, ECG, complication and integrating network. On average, this approach takes 175.833 milliseconds and 7.66667 epochs to complete the learning. The total training time for all networks to learn is 1055 milliseconds with 46 epochs.
Although many networks have to be constructed and trained separately, the multinetwork approach has reduced the complexity of network with large data set and has
overcome the limitation of the single network approach. This is because the networks represent all the possible combination of data, which were all used to train
them respectively. That is in the multi network approach all data sets are used in training. The knowledge (weight) produced by the network can be applied for all
possible data sets.

Item Type: Thesis (Masters)
Supervisor : UNSPECIFIED
Item ID: 979
Uncontrolled Keywords: Neural Network, Artificial Intelligence, Multi-Network Approach, Intelligent Medical System
Subjects: R Medicine > RC Internal medicine
Divisions: Faculty and School System > Sekolah Siswazah
Date Deposited: 15 Dec 2009 07:00
Last Modified: 24 Jul 2013 12:09
Department: Graduate School
URI: https://etd.uum.edu.my/id/eprint/979

Actions (login required)

View Item
View Item