Diabetes Classification using Deep Learning With a Hybrid Data Imbalance Handling Approach

Diabetes Mellitus (commonly known as diabetes) is one of the most prevalent chronic diseases worldwide, affecting millions of people. If not properly managed, it can lead to serious complications, including kidney damage, high blood pressure, heart disease, nerve damage, vision problems, and more. Therefore, correct classification from blood tests plays a crucial role. However, most of the available datasets contain imbalanced data such that normal (healthy) case samples are more (majority samples) than the unhealthy (diabetes) (minority) samples. In this work, we tackle this problem by introducing a hybrid data imbalance handling technique that focuses on minority samples. We test our balancing approach on different deep learning models such as deep neural networks (DNNs), Convolutional Neural Networks (CNNs), and a hybrid model. To assess the performance, we use the Pima Indian Diabetes dataset that contains 1-dimensional data about people, such as blood pressure, insulin, body mass index, etc. Experimental results indicate that our method improves the DNN model’s accuracy by 9.4%, F1-score by 26.8%, and AUC-ROC by 18.97%, on average, across five runs. Our class balancing approach provides promising results.