|
" Prediction of Diabetic Patient Readmission Using Hybrid Ensemble Learning "
Ghazo, Esraa
Khasawneh, Mohammad
| Document Type
|
:
|
Latin Dissertation
|
| Language of Document
|
:
|
English
|
| Record Number
|
:
|
1104914
|
| Doc. No
|
:
|
TLpq2279846127
|
| Main Entry
|
:
|
Ghazo, Esraa
|
|
|
:
|
Khasawneh, Mohammad
|
| Title & Author
|
:
|
Prediction of Diabetic Patient Readmission Using Hybrid Ensemble Learning\ Ghazo, EsraaKhasawneh, Mohammad
|
| College
|
:
|
State University of New York at Binghamton
|
| Date
|
:
|
2019
|
| student score
|
:
|
2019
|
| Degree
|
:
|
M.S.
|
| Page No
|
:
|
139
|
| Abstract
|
:
|
This thesis introduces the hybrid ensemble learning technique that combines and fully exploits different machine learning classifiers with their optimized hyperparameters such as SVM, GNB, PNN, MLP, DT, RF, KNN, LR and Keras (deep learning API). By using the output predictions for these classification algorithms as input for the ensemble learning algorithms such as boosting, bagging, voting. The output predictions for these ensemble algorithms also will be used as input prediction for the stacking method to produce a more robust model, give more efficient final prediction and achieve the best performance measures. Used on a binary classification dataset, this technique includes the combination of various first-level predictive models to produce a second-level model and then to produce a third-level model which leads to outperform all of them. Techniques such as dimensionality reduction using stepwise discriminant analysis, hyperparameter optimization using grid search, among others, have been adopted. This hybrid model is applied on the dataset before features selection and after features selection, using two techniques the genetic algorithm and filter algorithm. The hybrid model attained the highest accuracy, sensitivity, and AUC of 79.1%, 79.3%, and 81.3% respectively using 10-fold cross-validation with the genetic algorithm for features selection. T-test was also done with and without features selection to prove that the results are significant.
|
| Subject
|
:
|
Industrial engineering
|
| |