A Novel Solution of an Enhanced Error and Loss Function using Deep Learning for Hypertension Classification in Traditional Medicine

Abstract

Deep Learning in traditional medicine has different ways to detect and classify hypertension. However, not many researches have combined those ways to classify hypertension more accurately. This research aims to combine two of the most popular ways i.e. Tongue image and symptoms to increase the accuracy of detecting hypertension.The proposed system consists of training the parameters using error function with a Rectified Linear Unit (ReLU) Function and combining the learned features of both tongue image and symptoms using vector outer product. The proposed solution was tested on different data samples and provides the classification accuracy of 94.25% against the current average accuracy of 90.75%. The proposed solution only focused on increasing the classification accuracy. However, the proposed solution has not increased the processing time while doing so, instead the average processing time has decreased from 0.3774 to 0.3482.The proposed solution has increased the classification accuracy and decreased the processing time for classifying the hypertension in traditional medicine. The enhanced error function and loss function with ReLU activation function solves the vanishing gradient problem to achieve the accuracy of 94.25%.