Integrating Feature Ranking with Ensemble Learning and Logistic Model Trees for the Prediction of Postprandial Blood Glucose Elevation
Jason Chou-Hong Chen (Gonzaga University, USA)
Hsiao-Yen Kang (Landseed Hospital, Taiwan)
Mei-Chin Wang (Landseed Hospital, Taiwan)
Abstract: Postprandial blood glucose (PBG) elevation has been documented as a significant development of diabetes and cardiovascular diseases. Surprisingly, few studies have provided an effective model for predicting PBG elevation. This work presents the classification of PBG in a cohort study via integrating feature ranking with ensemble learning and logistic model trees. We used a cohort dataset that included 1,438 individuals from Landseed Hospital in Taiwan. Data from 2006 to 2013 were collected. To evaluate the performance of the proposed model, four well-known data mining classifiers (Naive Bayes tree algorithm, alternating decision tree, radial basis functions neural network, and Adaboost.M1) were employed in this study. The proposed model provided a reasonably accurate classification for predicting the PBG levels. Twenty-seven risk factors were identified as important risk factors for PBG elevation. The role of PBG should be emphasized and not that of PBG elevation. The predictive factors of PBG must be related to the development of certain diseases.
Keywords: chronic diseases, cohort dataset, data mining, postprandial blood glucose elevation
Categories: I.2.1, M.4