Keywords
Hospital Readmission, Type2 Diabetes, Healthcare, Machine Learning
Presentation Type
Talk
Research Abstract
Roughly 9.3% of US population suffer from diabetes and the 30 day readmission rate for diabetes patients range between 14.4 to 22.7%. Hence identifying the risk of readmission is a crucial information for the service providers to not only reduce the healthcare cost but also improve the quality of patient care. This paper models machine learning algorithms to compute probability of 30-day hospital readmission for type2 diabetes patients. Along with novel pre-processing techniques to identify the challenges of noisy and non-homogenized medical data, we used methods to downsize the feature vector size without sacrificing prediction accuracy. Our method has been implemented on a publicly available dataset from University of California Irvine at https://archive.ics.uci.edu/ml/datasets/diabetes+130-us+hospitals+for+years+1999-2008 which summarizes data from 130 US hospitals within span of 10 years with roughly 100,000 patients.
Session Track
Biomedical Engineering
Recommended Citation
Abhishek Sharma, Nan Kong, Paul M. Griffin, Andrés García-Arce, and Munirul Haque,
"Applying Machine Learning Techniques for Type2 Diabetes Readmission Prediction Based on Retrospective Data"
(August 2, 2018).
The Summer Undergraduate Research Fellowship (SURF) Symposium.
Paper 59.
https://docs.lib.purdue.edu/surf/2018/Presentations/59
Applying Machine Learning Techniques for Type2 Diabetes Readmission Prediction Based on Retrospective Data
Roughly 9.3% of US population suffer from diabetes and the 30 day readmission rate for diabetes patients range between 14.4 to 22.7%. Hence identifying the risk of readmission is a crucial information for the service providers to not only reduce the healthcare cost but also improve the quality of patient care. This paper models machine learning algorithms to compute probability of 30-day hospital readmission for type2 diabetes patients. Along with novel pre-processing techniques to identify the challenges of noisy and non-homogenized medical data, we used methods to downsize the feature vector size without sacrificing prediction accuracy. Our method has been implemented on a publicly available dataset from University of California Irvine at https://archive.ics.uci.edu/ml/datasets/diabetes+130-us+hospitals+for+years+1999-2008 which summarizes data from 130 US hospitals within span of 10 years with roughly 100,000 patients.