Abstract: A system capable of translating the logical activity into messages or commands for reactive applications also commonly called as Brain-Computer Interface (BCI). The project is divided in three phases. First Phase deals with Dataset collection, Binary class dataset analysis and Methodology (includes pre-processing of data, classification model, efficiency evaluation). Second phase discusses about complex dataset analysis (cognitive control, Prediction on finger flexion), Advanced algorithms for feature extraction and classification. Third phase discusses about Fine tuning of Algorithms, develop user Interface, Deploy ML model to interface. Different techniques like PCA (principal component analysis), Statistical Approach, Discrete Wavelet Transform (DWT) is used in extracting feature which is the basic techniques used in binary class datasets. Comparison and authentication between fundamental Machine Learning (ML) procedure using K- Nearest Neighbors KNN is done to give better prediction accuracy than the complicated Machine learning procedures like (Support Vector Machine (SVM), Artificial Neural Network (ANN), or Deep Neural Network (DNN). Results obtained for average efficiency for epilepsy detection using statistical and wavelet as features and SVM as classifier is 98%. Considering eye state detection dataset, obtained an average efficiency of 55%. Similarly for cognitive dataset an average of 95% efficiency was obtained by using PCA as a feature and KNN as classifier. In finger flexion dataset the aim was to find out the correlation between finger moments, by using the N best channel frequency pairs and applying the linear regression for the dataset, an average of 0.38 correlation was obtained. Thus, different method and classifiers have obtained fair predictions and efficiency was used. After training of ML models, the user interface was built using HTML and CSS and the trained models were connected to the interface by means of flask framework. The interface gives the user, freedom to select different combination of datasets and features and predict the class accordingly. The interface is also facilitated with visualization of the input test signal which uses chart for plotting graphs.

Keywords: BCI, Feature extraction, Classifiers, User Interface

PDF | DOI: 10.17148/IJARCCE.2021.101252

Open chat
Chat with IJARCCE