Dialysis is a cure for end-stage renal disease. Nephropathy is the leading factor that affects whether diabetic patients need dialysis treatment. In this study, a hybrid disease prediction scheme is proposed to identify critical disease risk factors for predicting diabetic patients who would suffer from nephropathy and need dialysis treatment in the future. The proposed model initially used under sampling based on clustering method to reduce the effect of class-imbalance problem of the real datasets collected from National Health Insurance Research Database (NHIRD). The C5.0 decision tree was then utilized to select important risk factors, and an extreme learning machine using the selected factors as predictors was applied to construct an effective disease prediction model to predict the diabetic patients suffering from nephropathy and needing dialysis. Empirical results revealed that the proposed hybrid disease prediction scheme not only provides better classification accuracy than that of the three competing models in terms of classification accuracy but also exhibits the capability of identifying important risk factors that can provide useful information for identifying high-risk groups for the dialysis of diabetic nephropathy. The results of this study provide an effective and appropriate hybrid disease prediction model to predict the dialysis of diabetic nephropathy.
透析治療是腎臟疾病末期的治療方法之一,而腎病變為影響糖尿病患者是否進入透析階段的主要因素。本研究提出一個整合式疾病預測模式來預測未患有腎病變的糖尿病患者於發生腎病變並進入透析治療階段及其相關的重要風險因子。在所提的方法中,首先透過集群減少多數抽樣技術來降低全民健康保險研究資料庫中常見的資料不平衡的問題。然後,使用C5.0決策樹技術來找出重要的風險因子。最後使用極速學習基於這些重要的風險因子來建構有效的糖尿病腎病變透析治療患者的預測模式。實驗結果顯示所提的整合式疾病預測模式不僅能產生優於其他三個比較模式的分類準確度,也能找出重要的風險因子,讓相關單位可對糖尿病腎病變透析治療之高危險族群加強健康管理,減少健保負擔。