Human Activity Recognition using Machine Learning Techniques

Human activity recognition (HAR) and gait analysis are two study topics that are used to identify numerous daily activities, such as walking, running, and stair climbing, and how they are performed. The valid identification of any gait deviation, as an abnormality in the gait cycle, can help in the real-time monitoring of patients with neuromuscular and musculoskeletal causes, and eventually in the restoration of their normal gait function. The current study combines multiple data preprocessing approaches with supervised machine learning algorithms to provide a framework for recognizing diverse gait activities using data samples from the publicly accessible “HuGaDB” human gait database. The automated analysis method takes into account 3-dimensional (3D) signals derived from two types of inertial sensors: accelerometers and gyroscopes, as well as electromyography (EMG) devices placed on the right and left leg of 18 healthy human participants. The proposed tool achieves a classification accuracy of 80% and Fl-score of 79% with Random Forest emerging as the optimal gait patterns identification method.

Paper

Full text

PDF

Human Activity Recognition using Machine Learning Techniques

Semantic Scholar · Computer Science · 2022

Abstract

Human activity recognition (HAR) and gait analysis are two study topics that are used to identify numerous daily activities, such as walking, running, and stair climbing, and how they are performed. The valid identification of any gait deviation, as an abnormality in the gait cycle, can help in the real-time monitoring of patients with neuromuscular and musculoskeletal causes, and eventually in the restoration of their normal gait function. The current study combines multiple data preprocessing approaches with supervised machine learning algorithms to provide a framework for recognizing diverse gait activities using data samples from the publicly accessible “HuGaDB” human gait database. The automated analysis method takes into account 3-dimensional (3D) signals derived from two types of inertial sensors: accelerometers and gyroscopes, as well as electromyography (EMG) devices placed on the right and left leg of 18 healthy human participants. The proposed tool achieves a classification accuracy of 80% and Fl-score of 79% with Random Forest emerging as the optimal gait patterns identification method.

Similar papers

© 2026 NYSGPT2525 LLC