Extraction and Motion Data Features of Wearable Devices Based on Genetic Algorithm Optimization

This paper proposes an online sequential extreme learning machine (OS-ELM) model improved using a genetic algorithm. First, wavelet threshold denoising was applied to eliminate sensor noise, and the sliding window method was used to segment the data. Second, a multi-objective fitness function was designed through the selection, crossover, and mutation operations of the genetic algorithm to dynamically optimize three key parameters: the number of hidden-layer nodes, the regularization coefficient, and the learning rate. Five statistical features, such as root mean square and zero-crossing rate, were extracted in the time domain, and three frequency-domain features, such as main frequency energy and spectrum entropy, were extracted using the Fast Fourier Transform. An improved OS-ELM model was then constructed for online sequence learning, and a dynamic forgetting-factor adjustment mechanism based on a genetic algorithm was introduced to enhance adaptability to time-varying data. Finally, model performance was evaluated under dynamic scenarios.