An epileptic seizures diagnosis system using feature selection, fuzzy temporal naive Bayes and T-CNN

Today's hospitals make use of state-of-the-art methods such as magnetic resonance imaging (MRI) and electroencephalogram (EEG) signal predictions in order to predict the occurrence of seizures well in advance. However, this method only works in theory. Almost every standard seizure prediction approach fails to predict epileptic seizures. Accurately, and most doctors mainly focus only on the treatment of epilepsy rather than preventing it. Health experts say that successive epileptic seizures would be even more virulent and fatal to individuals. The motivation of this research is to effectively predict and manage epileptic seizures disease by analyzing EEG signals. This work proposes a new Epileptic Seizures Diagnosis System (ESDS) for diagnosing epileptic seizures effectively. The proposed ESDS consists of two components, namely feature selection and classification. First, a newly proposed Fuzzy Temporal Naive Bayes (FT-NB) classifier and the existing Convolutional Neural Network with Temporal Features (T-CNN) are proposed for performing effective data preprocessing and classification. Second, a GridSearchCV method is used to determine the best parameter for hyperparameter tuning to obtain the best performance in the FT-NB. In addition, the T-CNN is also applied for enhancing the prediction result further, and the different machine learning (ML) algorithms are considered for performing comparative analysis with the FT-NB in terms of disease prediction. The experiments have been conducted by using the standard dataset and proved as better than other systems in terms of Precision (87.5%), Recall (90.4%), Specificity (97.4%) and Accuracy (96.7%).