Abstract: Gestational diabetes mellitus (GDM), which frequently endangers both the mother and her unborn child, is often detected between weeks 22 and 26 of pregnancy. However, the chatbot limitations include its dependency on static information and the absence of real-time data integration. It is also unable to adjust to varied demographics and does not offer personalized health feedback and continuous monitoring. In this paper, a novel DL-based DIA-CHAT approach has been proposed for diabetes classification using the Convolutional Neural Network-Bidirectional Gated Recurrent Unit (CNN-BiGRU). The input signals are preprocessed using the discrete wavelet transform (DWT) to remove the powerline interference and baseline wander from the ECG signal. The CNN-BiGRU integrates with a CNN to extract the shape and slope of the features and BiGRU is used to classify the diabetes and improve the performance. The CNN-BiGRU model classifies the ECG signals of pregnant women as diabetic and non-diabetic, and if diabetes is detected, the result is sent to a chatbot that provides personalized diet guidelines for pregnant women. The performance of the DIA-CHAT approaches was assessed using metrics such as recall, F1 score, accuracy, specificity, and precision. The DIA-CHAT maintains high accuracy levels of 99.67%. The DIA-CHAT approach enhances the total accuracy by 3.63%, 12.13%, and 1.52% better than OD-DSAE, En-RfRsK, and DiabChatbot, respectively.

Keywords: Diabetes, Convolutional Neural Network, Bidirectional Gated Recurrent Unit, pregnant women, ECG signal, discrete wavelet transform

Pages: 160-168