Thesis

Thesis Topic Details

Topic ID:
 3007
Title:
 Multi-instance Learning for Prediction of Days in Hospital
Supervisor:
 Alan Blair
Research Area:
 Machine Learning, Artificial Intelligence
Associated Staff
Assessor:
 Mike Bain
Topic Details
Status:
 Active
Type:
 R & D
Programs:
 CE BINF SE
Group Suitable:
 No
Industrial:
 Yes
Pre-requisites:
 COMP9417 or COMP3411
Description:
 In days of hospital prediction, historical patient health claims data are used to determine how many days the patient will spend in hospital in the following year. As each patient may have several claims, the problem is best represented as a multi-instance learning problem.
However, current multi-instance learning research commonly follow the standard multi-instance learning assumption. The standard assumption suggests that bag labels can be determined from a single within-bag instance, which is inappropriate for the particular problem domain. The thesis introduces a technique for multi-instance regression with relaxed assumptions, and demonstrate the feasibility and performance of using Gradient Boosting Machines to learn both instance level and bag level functions.
Comments:
 Would suit a student with good coding skills and a background in machine learning or artificial intelligence.
Past Student Reports
  Simon Jing-Bong KEUNG in s2, 2012
Machine Learning for Fraud Detection
 

Download report from the CSE Thesis Report Library

NOTE: only current CSE students can login to view and select reports to download.