Deep Learning For Improved Diabetes Detection: Integrating Lstm, Convolutional Neural Networks And Svm

Diabetes is one among the eminent anomalies which can be identified using ECG derived signal. This implies that if diabetes is not adequately managed with medications /insulin support it is likely to escalate into severe complicated diseases such as stroke, kidney failure, and heart attack. Early diagnosis of diabetes is also important in the management of the condition. The overall purpose of this thesis is to discover approaches to identify diabetes employing heart rate variability (HRV) signals as measured by deep learning such that to achieve very high level of correctness in the detection process. From the comparison of accuracies, CNN 1 - SVM begins at 68. 4%, while CNN 2 - SVM improves to 75. CNN 3 – SVM cast further has the highest accuracy of 93 percent. On the other hand, when we incorporate LSTM with SVM, the network configuration of CNN 1- LSTM - SVM results in an accuracy of 74. 2%, which is slightly higher than the CNN 1 –SVM. CNN 2 - LSTM - SVM further improves the accuracy to 76. 4% while and the highest accuracy is observed with CNN 3 - LSTM - SVM at 95.7%. In most cases, a combination of Long Short-Term Memory (LSTM) with Convolutional Neural Networks (CNN) and support vector machine (SVM) provides a better accuracy over the various models.