Fusing dimension reduction and classification for mining interesting frequent patterns in patients data

Abstract

Vast amounts of data are collected about elderly patients diagnosed with chronic conditions and receiving care in telehealth services. The potential to discover hidden patterns in the collected data can be crucial in making effective decisions on dissemination of services and lead to improved quality of care for patients. In this research, we investigate a knowledge discovery method that applies a fusion of dimension reduction and classification algorithms to discover interesting patterns in patient data. The research premise is that discovery of such patterns could help explain unique features about patients who are likely or unlikely to have an adverse event. This is a unique and innovative technique that utilizes the best of probability, rules, random trees and association algorithms for; (a) feature selection, (b) predictive modelling and (c) frequent pattern mining. The proposed method has been applied in a case study context to discover interesting patterns and features in patients participating in telehealth services. The results of the models developed shows that identification of best feature set can lead to accurate predictors of adverse events as well as effective in generation of frequent patterns and discovery of interesting features in varying patient cohort.