Purpose of Study: To classify and predict various complications related to diabetes by implementing data mining and supervised machine learning techniques and algorithms.Approach: This study used a dataset collected by the Rashid Center for Diabetes and Research, consisting of 884 records with 79 input attributes and eight diabetic complications: metabolic syndrome, dyslipidemia, neuropathy, nephropathy, diabetic foot, hypertension, obesity and retinopathy. Firstly, the dataset was processed to handle issues related to missing values and unbalanced data using three different methods (simple mean imputation, the k-NN model and the MissForest method). The next step was to encode categorical features in the dataset such as gender, nationality, and diabetes type, with a "dummy" variable. Data balancing was done using synthetic minority oversampling technique (for over-sampling the minority class) and cluster centroids (for under-sampling the majority class). Algorithms such as logistic regression, support vector machine, decision tree, random forest, AdaBoost, and XGBoost were extensively tuned and trained to build the actual model. To select the best hyperparameters, grid search with cross-validation was employed. Further, to split the dataset for training and testing, k-fold cross-validation with k = 10 was utilized. The performance of the built models was evaluated using classification accuracy, precision, recall and Fl scores.
展开▼
