Leveraging Wearable Sensors and Supervised Learning Paradigm as a Configurable Solution for Epileptic Patients

Abstract

Epileptic seizures are among the most frequently occurring and unpredictable chronic neurological disorder that disrupt the lives of affected individuals. Thus, it paved the way for including Machine and Deep Learning models in the present frameworks for intelligent, self-driven epileptic seizure management. The few commonly deployed methods are Electroencephalogram (EEG), Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and Electrocardiography (ECG). However, low amplitude and fluctuations make it difficult for ML algorithms to achieve satisfactory results in ambient, harsh environmental conditions. Moreover, several proficient models, such as CNN and Random Forest, take excessive computational time in the training phase of the program. Furthermore, EEG hampers the flexibility of patients by its monitoring procedure confined to one room. Moreover, techniques like Auto encoding face issues of false negative rates (FNR). The paper presents a novel and robust framework using wireless sensors, with an increased data points for a competent KNN algorithm. The model demonstrated is compatible with the patient's daily routine activities and can predict the frequency of seizures with a 2.2% error rate. Instead of using 5-22 subjects as in prior studies, the algorithm is applied under 32 patients, which optimizes its performance rate. The practice fostered durability of the model by preparing it for various unpredictabilities. This paper also presents a comparative overview of the novel paradigm with the current systems based on accuracy rate and dataset size. It also sheds light on the limitations of presently deployed architectural configurations and presents a sustainable solution for the need for a pliable and credible epileptic monitoring regime.