Detection of Epilepsy Using Adaptive Neuro-Fuzzy Inference System and Comparative Analysis

This study presents the use of Adaptive Neuro-Fuzzy Inference System (ANFIS) for classification of the EEG signals. The data consists of two types of EEG signals, i.e. epileptic patients during epilepsy and healthy patients when their eyes are open. We propose two algorithms for the detection of epileptic patients. In the first algorithm we use Discrete Wavelet Transform (DWT) and statistical analysis for feature extraction, whereas Principal Component Analysis (PCA) is used in order to reduce the number of features in the second algorithm. ANFIS model learns how to classify the EEG signal, through the standard hybrid learning algorithm, whereas we use special form of ANFIS model, which depending on the number of inputs, splits the model into appropriate number of substructures (sub-ANFIS models). The algorithms were evaluated in terms of training performance and classification accuracies. From the simulation results it was concluded that the both algorithms have good potentials in classifying the EEG signals. Further, a comparative analysis for the influence of the tuning parameters was made, i.e. the influence of the different data splitting methods, the influence of the different input space partitioning methods, the usage of the different wavelet functions in the WT, the effects of normalization, as well as the effects of using different membership functions. From the analysis it was concluded that different combinations of input parameters differently classify the EEG signals. Lastly, a comparison of the both algorithms was made, in terms of training performance and classification accuracies, whereas it was concluded that the algorithm that uses PCA for feature extraction, in some cases, performs better than the algorithm that uses DWT, even though the number of features is significantly reduced (from 20 to 7).