Implementation of a Hybrid Approach for Chronic Disease Risk Assessment and Recommendation System

Abstract

There are a number of methods that have been developed by researchers in an attempt to predict and diagnose chronic disease conditions accurately. These methods play an important role in controlling the conditions of those who suffer from such diseases. Furthermore, they can assist healthcare providers to have 24/7 remote monitoring system when integrated within an ehealth management system on the condition that these methods provide an accurate and trustworthy result. These methods can assist patients to have 24/7 access to medical care as well as monitoring their conditions. Within this research a healthcare recommender system for chronic disease risk assessment has been advised. The system proposed in this research serves three purposes; participating and assisting physicians for monitoring and predicting condition of their patients remotely, helping patients to have access to healthcare providers irrespective of time and location, and educating patients towards controlling their conditions. Medical data are complex in nature and are usually made of a large, multidimensional, unbalance as well noise and missing some of the instances that make the full health record. Hence, the task of advising an accurate method is very challenging. Therefore a new recommender system is developed in this paper that is based on a combination of two methods in order to create a hybrid solution. The new approach is made of using both Multiple Classification ( MF ) and a unified Collaborative Filtering ( CF ). MF is adopted to build an accurate model that is able to assess, predict, and diagnose monitored cases with high accuracy. Whilst, the Collaborative Filtering ( CF ) is implemented to achieve an accurate recommendation when providing a medical advice to patients. The latter method ( CF ) is based on learning classification model using both historical binary data and external features that are highlighted in this paper. This research is conducted in the United Arab Emirates ( UAE ) where historical medical data were obtained from the region and formed the base model that is used for both prediction and providing advice and recommendations. In order to reduce the vast amount and to reduce the complexity of medical data as well as eliminating those attributes that have none or negligible effect on diagnoses, then only relevant attributes that are used to create electronic medical records were formed. Records were made of a combination of static and dynamic data. Both of these types are combined to generate a health record that leverage diagnoses reliably.