Human hearing disorder recognition model using eeg-aep based signal

Hearing deficiency is the most prevalent sensory impairment worldwide, impeding human learning and communication. The appropriate technique for dealing with this concern is an early and accurate hearing diagnosis using an electroencephalogram (EEG). The most significant modality for diagnosing hearing deficiency among EEG control signals is the auditory evoked potential (AEP), which is generated in the cortical region of the brain through auditory stimulus. This study aims to build an intelligent system for hearing deficiency diagnosis. Firstly, fifteen feature extraction methods were used to extract information from the raw AEP features. Then, the extracted feature was classified using machine learning and deep learning algorithms. The performance of the proposed approach was validated using the experimental collected dataset (UMP-Emotiv-AEP) and well-known publicly available AEP datasets. In this study, a deep learning model (improved-VGG16) was designed for detecting hearing deficiency. To improve the VGG16 architecture, some layers of the based VGG16 model were replaced with new layers in the fully connected block. Additionally, during the training of the proposed model, some convolutional layers of the base VGG16 model were frozen where the model used the pre-trained weights. Next, the fine-tuning technique was used for the remaining layers to fit the dataset with the proposed improved-VGG16 model. This study investigated two conventional machine learning algorithms: support vector machine (SVM) and k-nearest neighbors (KNN), and two deep learning techniques: convolutional neural network (CNN) and improved-VGG16 model. Among these, the proposed continuous wavelet transforms (CWT) with improved-VGG16 architecture showed outstanding performance for hearing deficiency diagnosis. Here, the CWT was used to convert raw AEP signals into time-frequency images. Two types of AEP signals were collected using a five-channel Emotiv insight device (UMP- Emotiv-AEP), recorded from ten subjects. Three experimental analyses were conducted with the UMP-Emotiv-AEP, the proposed architecture showing a significant improvement with 99.90% testing accuracy. The proposed architecture achieved a 2.09% and 0.32% improvement in accuracy compared to the KNN, and CNN approaches. Secondly, a popular publicly available dataset was used, collected from sixteen subjects. Three experimental analyses were conducted with the publicly available dataset, whereas the proposed architecture outperformed the state-of-art studies by improving the classification accuracy to 96.87%. The proposed architecture achieved a 1.58% and 1.66% improvement over the performance of the SVMs approach. The experimental outcomes demonstrated that traditional machine learning and CNN algorithms achieved comparatively lower accuracy than the proposed model. Additionally, this study developed a user-friendly graphical user interface (GUI). The proposed approach’s performance indicates that it can significantly deal with AEP response for hearing deficiency diagnosis