Detection of Diabetes Using 1D Convolutional Neural Network from Short Time PPG Signal

Diabetes is a prevalent and chronic condition requiring effective surveillance and categorization. In recent years, the analysis of physiological signals, such as photoplethysmography (PPG), has arisen as a promising detection and classification technique for diabetes. The main focus of this study is to analyze the detection of Diabetes using 1D Convolutional Neural Networks (CNN) applied to short-time PPG signals. Our experimental findings demonstrate that the proposed method is effective. Using 5-fold cross-validation to evaluate a dataset containing PPG signals from diabetic and non-diabetic subjects, we attain an average accuracy of 84.62%. In addition, we report precision, recall, and F1 scores to demonstrate the model's ability to correctly classify diabetic and non-diabetic. For the ‘diabetic’ class, our method achieves a precision of 0.864041, recall of 0.857135, and F1 score of 0.868085. For the ‘non-diabetic’ class, our precision is 0.83114, recall is 0.858586, and F1 score is 0.845771. These results demonstrate that our 1D CNN-based method accurately classifies Diabetes from PPG signals, allowing for early detection and intervention.