EEG Feature Prediction from Tactile Data to Improve Object Shape Classification

In this work, we analyse the Electroencephalogram (EEG) and tactile signals acquired during dynamic exploration of objects of seven different geometric shapes and observe that classification performance using features from both the domains together is better than using the either alone. Classification is done by Support Vector Machine and Naïve Bayesian (NB) classifiers using discrete wavelet transform features. ReliefF algorithm is implemented for feature dimension reduction. A 6th order polynomial is fitted to tactile features to predict the EEG features which is helpful in cases where EEG data is unavailable. These predicted features recognize object shapes with improved classification accuracy when used with tactile features than using either of them separately. The results depict that object shape recognition rate using Naïve Bayesian classifier has been enhanced from 75.28% in case of tactile features to 82.63% for dimension reduced tactile features along with predicted EEG features.