| dc.description.abstract | Diabetes disease is a chronic disease whose prevalence continues to increase throughout the world. Diagnosis of this disease can be done through blood sugar checks and Glycated Hemoglobin (HbA1C) tests, as well as Complication Screening Tests. Several previous studies have succeeded in creating a diabetes detection system using algorithms Random Forest, Naive Bayes, and so on, but this research uses algorithms Deep Neural Network (DNN) and algorithms Support Vector Machine (SVM) by using dataset publicly available on Kaggle. This approach involves data pre-processing,techniques oversampling data, division of data into training-validation-testing sets, data normalization, and use of algorithms Deep Neural Network (DNN) and algorithms Support Vector Machine (SVM) in model development machine learning. The DNN and SVM architecture used consists of layers input, followed by 3 dense layer with 32, 16, and 8 neurons respectively, and using the ReLU activation function (Rectified Linear Units). There are also 3 layers dropout with a rate of 0.4 between each dense layer, and layers output with activation function sigmoid. This model is compiled with optimizer Adam with a learning rate of 0.001, using functions loss squared hinge, and accuracy metrics. Model training is performed for each dataset original, dataset with DNN algorithm, SVM algorithm, and DNN-SVM algorithm. The results show the best accuracy in dataset The DNN-SVM algorithm was applied with a training accuracy of 97,12% and loss training was 0,4279, with validation accuracy of 97,12% and loss validation of 0,4256, and testing accuracy of 97,12% and loss testing of 0,4256. This model is then integrated into website to make it easier for users to detect whether they suffer from diabetes based on the data entered. | en_US |
