Muhammad Zaisya Fitriannuur Rahman, Dodon Turianto Nugrahadi, Mohammad Reza Faisal, Andi Farmadi, Muhammad Itqan Mazdadi, Irwan Budiman, Favorisen Rosyking Lumbanraja, Umiatin
Diabetes is a chronic disease that requires early and effective detection to prevent long-term complications. Conventional diagnostic methods remain invasive, often causing discomfort, limited accessibility, and dependence on medical personnel. This study implements a deep learning approach to non-invasively detect diabetes through the analysis of electrocardiogram (ECG) signals. Three model architectures - Convolutional Neural Network, Long Short-Term Memory, and Gated Recurrent Unit - are evaluated using four variations of input shapes. ECG signals are processed through filtering and segmentation stages based on R-peak detection to obtain representative inputs. The best result is achieved by the CNN model with an input configuration of 500 rows and 2 features, reaching an accuracy of 94.00%. These findings indicate that input shape significantly affects the performance of sequence-based models, while CNN remains consistent across various configurations. This research advances the field of early diabetes prediction by systematically evaluating different neural network architectures and input signal representations. © 2025 IEEE.
Lambung Mangkurat University, Department of Computer Science, Banjarbaru, Indonesia; University of Lampung, Department of Computer Science, Lampung, Indonesia; Jakarta State University, Department of Physics, Jakarta, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock