Deep Learning Based Cardiovascular Disease Risk Factor Prediction Among Type 2 Diabetes Mellitus Patients


  • C. Selvarathi Department of Computer Science and Engineering, M.Kumarasamy College of Engineering, Karur, Tamil Nadu, India-639113
  • S. Varadhaganapathy Department of Information Technology,Kongu Engineering College, Erode, Tamil Nadu, India-638060



Cardiovascular disease, Type2 Diabetes Mellitus, Deep learning, cascaded, Long short term memory, Convolution, Evaluation metrics


Type 2 Diabetes Mellitus (T2DM) is a common chronic disease that is caused due to insulin discharge disorder. Due to the complication of T2DM, the outcomes of this disease lead to severe illness, death and cardiovascular disease (CVD). Given a larger number of diabetes patients, it is necessary to find the patients with a high risk of CVD complications. For this, the traditional methods are not sufficient and it is important to develop a deep learning-based efficient quantitative model to predict the risk of CVD among diabetes patients. The major objective of this research is to assess the efficient artificial intelligence approach toward the proposal of a personalized deep learning model that can able to predict the risk of fatal and non-fatal CVD among T2DM patients. First, the unbalanced dataset is preprocessed to make the dataset balanced for processing. Second, the features are reduced and important features are selected using Rank based Feature Importance (RFI) model which will improve the prediction accuracy. Third, the proposed Cascaded Convolution Graph LSTM (CCGLSTM) has been used as a classifier to predict the risk of CVD. Novelty of the work resides on ranking based feature analysis is cascaded with CGLSTM. The proposed model is implemented and experimented with various evaluation metrics using the data from 560 patients of five-year follow-up with T2DM. These evaluated results are compared with the state of-the-art methods and the proposed model is proven to be superior to other approaches in terms of AUC (0.989),
Accuracy (98.8%), recall (96.7%), precision (96.8%), specificity (97.4%) and F1-Score (97.5%).