Stability Analysis for Bipedal Robots using Hybrid Deep Learning Techniques for Effective Human Activity Recognition

Abstract

This research endeavor proposes an innovative approach to enhancing human activity newline newlinedetection by utilizing smartphone sensor data. Smartphones are perfect tools for ac- newlinetivity detection because of their widespread use and integrated inertial sensors, such as newline newlineaccelerometers. To correctly classify actions including running, walking, sitting, stand- newlineing, stair climbing, and stair descending, the suggested approach uses a stacked ensem- newlineble learning architecture. The experimental results show that the CatBoost classifier newline newlinehas an accuracy of 80.41% for discriminating between walking and running, 82.34% newline newlinefor stair-related activities, and 98.78% for distinguishing between sitting and stand- newlineing. These results show a significant improvement in categorization performance. The newline newlineStacked Ensemble Learning (SEL) methodology is extremely effective in healthcare newlineand other sectors, outperforming conventional methods. This improvement is further newlinesupported by cross-validation and confusion matrix analysis. Future study could look newlineinto improved ensemble models to reach even more accuracy. newline