Machine Learning-Driven Robust Optimization of Communication Signals in Sensor Wearable Devices for Early Stage Epilepsy Seizure Prediction using EPCA

Abstract

Objectives: To introduce a novel EEG signal optimization and epilepsy seizure detection method at an early stage with the aid of sensor wearable devices in order to treat the epilepsy in advance. For effective optimization and to boost the detection accuracy, the ROCS-EDS (Robust Optimization of Communication Signals for Early Detection of Seizures) technique is employed with EPCA (Enhanced Principal Component Analysis). Methods: EEG signal optimization, feature selection, and extraction such as time-domain, frequency-domain, time-frequency, movements, and statistical features are done by EPCA. IQAM and CPSK techniques are utilized for data transfer and reliability. The BCI C-IV (Brain Computer Interface Dataset from California) dataset is used in this study with a record of 4922 patients, a 400 Hz sample rate, 4 target classes, and 2 subjects. LD parity check, fast Fourier, and wavelet transform methods are deployed to track signals and optimize them at frequent intervals. For filtering the signals and reliability rate monitoring, the MMSEF error filtering method is employed. MATLAB software is used to measure the performance of the suggested model with baseline models such as VPSOGA-SVM, 2L-LSTM, and IoMT-ESD. Findings: With an accuracy rate of 98%, 96% and 97% sensitivity and specificity, a 98% F-score, 98.65% and 96.80% precision and recall, and 97% and 91% TPR and TNR, 5.46% and 7.10% FPR and FNR, ROCS-EDS outperforms baseline models in terms of signal optimization and seizure detection at an early stage. The method is deployed and tested in MF2Epi-alert wearable device for seizure activity. Novelty: The novel hybrid model ROCS-EDS with EPCA shown evident results and has capability in optimizing EEG signals and detect ES accurately at an early stage which helps the clinical experts to treat the epilepsy in advance. ROCS-EDS outperform the currently prevailing methods VPSOGA-SVM (1) , 2L-LSTM (2) , and IoMT-ESD (3) .