Abstract
Mental state of a person is not easy to read by humans. This mental state may include anxiety, depression and other medical conditions. There is no correct physical testing that is available. Therefore, diagnosing these mental conditions in yearly stage is very important. In our proposed model, we use EEG signal from non-intrusive wearable devices to detect depression patterns in patients. This can be done with the help of proposed Hybrid CNN (Convolutional Neural Network) algorithm. In this algorithm we are introducing adaptive wavelet kernels in convolutional layers to reduce the number of connected neurons in the network. This Mental state diagnosis using Hybrid CNN with EEG signal algorithm is also designed to be shift invariant and can handle real world online data from wearable devices. Finally, the training time requirement of neural network can be reduced with the help of improved learning rate functions. This kind of improvements will make the algorithm consume less training time and power. The algorithm has an accuracy of about 97% from data that is taken from wearable device with very poor SNR ratios. The main motive of the paper will be to construct fast, reliable, energy efficient algorithm for real world wearable applications.