An efficient empirical mode decomposition based feature extraction model for human activity recognition of elderly people using machine learning algorithms

Abstract

Many individuals throughout the globe need to be constantly monitored for health reasons, including diabetes patients and individuals with other chronic diseases, the elderly, and the disabled. At any time, these individuals may be at a higher risk of suffering life-threatening falls or experiencing fainting. HAR (Human Activity Recognition) model using machine learning techniques play an important role in observing the activities of the people. The existing methods of activity monitoring lacks accuracy. Hence, the proposed method focuses to improve the prediction accuracy using accelerometer and gyroscope data. The research work analysis accelerometer and gyroscope data using various decomposition techniques such as EMD(Empirical Mode Decomposition), DWT (Discrete Wavelet Transform), FFT (Fast Fourier Transform) to process non-linear data and to split series of signal data into set of IMF(Intrinsic Mode Function), PCA(Principal Component Analysis) was performed for selecting optimal features. Then human activities are recognized by using multi-class classification techniques. The proposed EMD method achieves better performance with 98.4% accuracy, 100% Precision, 100% Recall and 100% F-measure.