Abstract——The invasive glucose measurement and classification methods offer high diagnostic accuracy; however, they have several significant drawbacks, including patient discomfort and elevated costs due to the frequent use of glucose test strips. Consequently, previous research has explored non-invasive alternatives, which are more convenient and cost-effective. A prominent non-invasive approach involves the use of Photoplethysmography (PPG) signals processed with deep learning models to estimate blood glucose levels and classify diabetic status, but it still faces challenges in achieving optimal diagnostic accuracy. Therefore, this paper proposes a diabetic classification framework and evaluates the performance of CNN models in classifying diabetic and non-diabetic subjects using PPG signals. The research method employs a computer simulation, starting with the modification of PPG signals into five distinct datasets using a moving average filter that varies the number of points (M). The CNN models' performance is evaluated using transfer learning with pre-trained weights from ImageNet, utilizing architectures from the ResNet, EfficientNet, and MobileNet series. Each model is trained and tested separately to classify diabetic and non-diabetic. The results indicate that the size of the moving average window M is a crucial factor in determining model performance. Utilizing a dataset filtered with an average of M = 10 points, all evaluated architectures—including ResNet (34, 50), EfficientNet (V2-M, Lite), and MobileNet (V2-S, V3-L)— achieved perfect classification performance, yielding 100% accuracy, precision, recall, and F1-score across all test parameters. These findings indicate that the proposed model with M = 10 can effectively classify individuals as diabetic or non-diabetic, providing a promising non-invasive alternative for glucose monitoring.

Index Terms—convolutional neural network, deep learning, diabetes, non-invasive, photoplethysmography signals, moving average filter

Cite: Zulhelmi Zulhelmi, Roslidar Roslidar, Sawal H. Md Ali, and Nasaruddin Nasaruddin, "Evaluation of CNN Models Based on Photoplethysmography Signals Using Optimal Moving Average Filter for Non-Invasive Diabetic Classification," International Journal of Electrical and Electronic Engineering & Telecommunications, vol. 15, no. 4, pp. 288-303, 2026. doi: 10.18178/ijeetc.15.4.288-303

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).