<![CDATA[Diabetes Mellitus (DM) is a chronic disease characterized by blood glucose (blood sugar) levels exceeding normal, i.e. blood sugar levels being equal to or more than 200 mg/dl, and fasting blood sugar levels being above or equal to 126 mg/dl. The increase in the number of people with diabetes is due to delays in detection. Utilization of machine learning in helping to establish a fast and accurate diagnosis is one of the efforts made in the health sector. One of the important steps to produce high classification accuracy is through the selection of relevant features. The problem in feature selection is dimensionality reduction, where initially all attributes are required to obtain maximum accuracy while not all features are used in the classification process. This study uses the Binary wheal Optimization Algorithm (BWOA) as a feature selection method to increase accuracy in the classification of diabetes mellitus. The use of metaheuristic algorithms is an alternative to increase computational efficiency and avoid local minimums. The BWOA algorithm reduces the 8 attributes in the dataset to the 3 best attributes that are able to represent the original dataset. The results showed that from the six classification methods tested, namely: K-NN, Naïve Bayes, Random Forest, Logistics Regression, Decision Tree, Neural Network. then the three logistic regression methods, naive Bayes and neural network are in good classification criteria based on Area Under Curve (AUC) while the calculation of the accuracy value shows an average of above 70%. Keywords : Feature Selection, Classification, Diabetes Mellitus, Accuracy, Area Under Curve (AUC)]]>
Copyrights © 0000
