A Tri-Fusion Approach for Brainwave-Based Biometric Authentication Using Consumer EEG Devices

Brainwave biometrics holds promise for secure authentication, but its real-world adoption faces challenges. Most existing models are tested on small datasets, which include less than 50 participants. In addition, many studies either analyze complete brainwave patterns or concentrate on a single wave associated with rest or sleep, which makes them impractical for authentication. These limitations restrict the real-world use of brainwave biometrics. To address these challenges, this paper proposes a novel brainwave-based authentication model known as 'TFG-SVM' utilizing the most important brainwaves to set a step forward for futuristic biometrics using advanced consumer EEG devices. This model uses a Tri-Fusion framework in coordination with a Support Vector Machine (SVM) and k-nearest Neighbors (k-NN) algorithm. i) First, we begin with the Early Fusion in coordination with k-NN algorithms to develop a graph-based representation of local relationships between brainwave signal instances across both spatial and temporal dimensions to facilitate nuanced learning processes within the GNN layers. ii) Next, we apply late fusion to combine outputs from separately processed signal channels at the decision level. iii) We then use weighted fusion to dynamically optimize the contribution of each pathway based on their predictive reliability, which significantly improves the accuracy of the model. iv) Finally, a Support Vector Machine (SVM) classifier refines the decision boundary, effectively distinguishing between authentic users and imposters by maximizing the margin between them. We evaluate our proposed model on different datasets and compared its results with state-of-the-art methods and other algorithms implemented in this work. This comparison shows how effectively our model ensures legitimate user authentication by achieving high accuracy approximately 97.9%, and making it suitable for real world applications.