ipl-logo

Readmission Classification Model

938 Words4 Pages

years 1999-2008” dataset as training dataset and 15% of the whole dataset as test dataset. To evaluate the effectiveness of the proposed RF based readmission classification model, we validated the proposed model using the test subset of the Diabetes 130-US hospitals for years 1999-2008 dataset. We applied 10 - fold cross validation strategy on training dataset, and then the cross validated model was tested on the validation dataset. Each of the experiments was run for multiple times with different random seeds, and the results were achieved by taking mean over different experimental runs. In this work compared the proposed model against two classic models such as Naïve Bayes, and C4.5. We displays the comparison in different methods. In this …show more content…

After that we classify the patients who are readmitted within 30 days of their release from hospital from who are readmitted after 30 days of their release from hospital within the previously classified readmitted patients. Therefore, we applied random forests on the selected 47 features in which we set the ensemble size ntree was equal to 600, and 1000 in the case of classifying readmitted patients and in the case of classifying readmission within 30 days from readmission after 30 days, respectively; and sub-space dimensionality mtry was selected by applying grid search over the possible values of …show more content…

The produced results from random forests in the case of classifying after 30 days of release was 6.8% and 3.5% better than C4.5 and naïve Bayes classifiers in terms of the AUC metric. On the other hand, in the case of predicting the readmission within 30 days of release, all of the three classifiers low accuracies around 29-40%.
Therefore, we can conclude that the degree of difficulty of predicting whether a patient will readmitted within 30 days of releasing from hospital was very high. As we achieved this classification results using some meta-features about patients, diagnostics, and medication, we can therefore utilize this results for further improvement of the patient monitoring procedure, medical practice, and minimization of the expense of

Open Document