Incrementing FRNN model with simple heuristic update for brainwaves person authentication

This paper proposes an incremental learning approach to classify EEG signals for person authentication modelling. Biometric application receives only limited personal training data samples in practice. The knowledge granules that represent individual's biometric characteristics need to be updated from time to time, to ensure the authentication performance. This is even crucial when the initial personal data samples are not in good quality, i.e. contain noise. Hence, incremental update method in the proposed authentication model aims to reform the knowledge granules incrementally via model tuning. The proposed Incremental Fuzzy-Rough Nearest Neighbour (IncFRNN) algorithm is embedded with a heuristic update method which is able to reshape and reform the personalized knowledge granules through insertion and deletion of a test object. The proposed incremental update function updates separately on the personalized knowledge granules, based on similarity measures and predefined window size upon undesired performance. The proposed IncFRNN model was compared against its predecessor, the conventional FRNN model, to evaluate its performance for person authentication modelling using EEG signals. The results show that our proposed IncFRNN model has successfully yielded better results in terms of AUC.