Efficient Classification and Wireless Communication for Electrooculography Signals Based on Internet of Things and Machine Learning (IoT-ML) Algorithms

Abstract In this paper, we investigate an Internet of Things (IoT)-based platform utilizing Electrooculography (EOG) to assist, control, and monitor a smart home environment in real-time for patients with motor disabilities. Users can interact with the intelligent environment through a Graphical User Interface (GUI) that offers predefined options for controlling doors, windows, lights, air conditioning, temperature, and TV functions. The proposed approach is based mainly on utilization of two transforms namely Stockwell (S-transform) and Wavelet transforms respectively for detection of abrupt changes EOG signals. Several signal statistical features including maximum, minimum, mean, median, Root Mean Square (RMS), standard deviation, Zero Crossing Rate (ZCR), Mean Curve length (MCL), kurtosis and skewness of the processed signals are utilized to characterize the EOG signals which have been applied for the classification stage to detect one of eye movement directions including: up, down, right, left, no movement or blinking. Two different wavelet families including Daubechies (db4) and Symlets (Sym4) wavelets are considered. The proposed approach was conducted on a data set which acquired by a custom device to measure and record EOG signals. Then, the decomposed horizontal and vertical signals are grouped into a vector to be the inserted as an input to the classification models. Finally, the data are classified using three types of Machine Learning (ML) algorithms including Support Vector Machine (SVM), Kernel Neural Network (KNN) and Ensemble Tree (ET) classifiers. The result of the proposed method outperforms the results of the other previous published methods when using different evaluation metrics. A high average accuracy of 97.7% is achieved from SVM using db4 which demonstrate the superior performance of the proposed method