Diabetes Readmission Prediction using Distributed and Collaborative Paradigms

Analyzing enormous amounts of healthcare data to obtain meaningful insights requires efficient and timely solutions. Diabetes is one of the most critical chronic healthcare problems that affect other organs of the human body. Hospital readmission, for patients with diabetes, is a common scenario where a discharged patient is admitted again within a specific time interval. Efficient techniques are needed which can predict the chance of such a readmission, thereby, allowing the possibility of targeted interventions. The aim of this paper is to discuss the performance of different prediction algorithms and associated collaborative paradigms for publically available diabetes data. Apache Spark is used, in the prototype, to decrease the training time. The prototype also addresses underlying challenges such as fault tolerance, scalability, and heterogeneity. The results of various experiments show that the collaborative technique increases the accuracy of a poor performing prediction algorithm by around 22% in one collaborative configuration.