估計有近七成五的國人罹患二種以上的慢性疾病，而糖尿病與慢性腎臟病經常列在國人十大疾病中。糖尿病是引起慢性腎臟病最主要的原因，容易導致各種併發症及增加死亡率。若能及早預測腎臟功能提供醫療照護，必能延緩末期腎臟病的病程。綜觀探討糖尿病人預測慢性腎臟病的相關研究，監督式機器學習以隨機森林演算法具有優異的成效。本研究利用第二型糖尿病人之口服降血糖藥品類別與生化檢驗資料，利用隨機森林演算法建立預測模型。本研究有別於既往之研究，預測模型以藥物治療分類為主要特徵維度，經彙整藥學相關專業知識，對降血糖藥物進行適當的特徵選取，並運用皮爾森相關分析於生化檢驗中篩選適當之特徵項，去除雜訊的特徵值，歷經參數調整後，其預測結果的F1−Score值可達到約82.9%。

There are nearly 75% of people have more than two chronic diseases in our country. Diabetes and chronic kidney disease are often among the ten top diseases. Diabetes is the main cause of chronic kidney disease which various complications and increases mortality. If we can predict kidney function early, that will provide the treatment decision for medical caregivers. Through the literature review, random forest algorithm has relatively excellent results among supervised machine learning which predict renal function with diabetic patients. This study attempts to screen the features from the data of hypoglycemic agents and biochemical test data. This study is different from previous studies which the main feature selections are based on biochemical test values and international classification of diseases codes. This study performed the anatomical therapeutic chemical classification system as the main feature dimension. After selecting appropriate features through the domain knowledge of pharmacology, we apply the random forest algorithm for model training. The F1_Score of prediction model can reach nearly 82.9% after adjusting the parameters.