Fig. 4: Eigengraph deep learning analysis process.
From: Data-centric artificial olfactory system based on the eigengraph
a Preprocessing flow for extraction of 20-dimensional MFCC feature vectors based on fast Fourier transform results. b Deep neural network with 3 hidden layers activated with ReLU function. c 3D confusion matrix indicates the number of correct prediction to confirm confusable sets between the 117 predicted and true labels when the training accuracy was temporarily 90%. d, e Training accuracy and training loss graphs according to the training sample size using learning rate of 0.0001 and 0.00001, respectively.
