Ανάπτυξη μοντέλων μηχανικής μάθησης στο Διαβήτη
Date Issued
October 19, 2023
Type
Πτυχιακή Εργασία
Abstract
This thesis focuses on the construction of two machine learning models for predicting the hospital readmission rate and length of stay of diabetic patients. The dataset used is the "Diabetes 130-Hospitals Dataset" and includes clinical information on more than 100,000 diabetic patients from 130 hospitals. After preprocessing the data, the optimal features were selected using the WSAE and CSAE methods for the classification and regression problem respectively. Eight algorithms were used for the classification problem with Voting (NB, LR, DT), Bagging (LR) and Stacking (LR) being the 3 best algorithms where the first algorithm performed well for the class "<30", the second algorithm for the class "No" and the third algorithm for the class ">30" while all of them had similar average performance. For the regression problem, 3 different algorithms were used, where the K-Nearest Neighbors Regressor (K-NNR) algorithm showed the best performance with the lowest mean absolute error (MAE) and root mean square error (RMSE) values. Furthermore, a web application was implemented for the use of the above models by diabetic patients and health professionals. In conclusion, it turns out that the use of machine learning turned out to be quite beneficial as it not only helps in the personalized and early care of diabetic patients by preventing readmissions or death rates due to delayed care but also helps in the better allocation of financial resources of hospitals.
Subjects