Researchers created an AI-powered wearable framework that identifies 24 human activities with 98% accuracy, supporting reliable health monitoring, rehabilitation and movement analysis applications.
Researchers have developed a new artificial intelligence framework using a single hip-mounted wearable sensor, achieving nearly 98% accuracy in recognising 24 different human activities. The study, published in Scientific Reports, combines convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) networks and an attention mechanism to improve the accuracy of activity recognition from motion sensor data.
The framework analyses data collected by accelerometers and gyroscopes embedded in a hip-mounted wearable device. Researchers processed eight motion inputs, including three-axis acceleration and angular velocity, before applying a lightweight CNN to extract local movement features. These features were then analysed by a two-layer BiLSTM network, while an attention mechanism highlighted the most relevant motion patterns and reduced interference from irrelevant signals.
The model was evaluated using nearly 400,000 segmented sensor windows and achieved almost 98% accuracy, precision, recall and weighted F1-score under standard testing conditions. It successfully recognised activities including walking, running, jumping, using lifts and different movement directions. However, researchers noted that distinguishing between similar static postures, such as sitting and standing, remained more challenging.
To assess real-world performance, the framework also underwent leave-one-subject-out cross-validation, where each participant was excluded from training in turn. Under these conditions, the system achieved an average accuracy of 78%, reflecting natural differences in body shape, gait and movement patterns between individuals. Even so, the results compared favourably with conventional CNN and recurrent neural network models, while requiring only ten training epochs to converge.
Researchers believe the technology could strengthen wearable healthcare by supporting fall detection, rehabilitation monitoring, remote patient care and chronic disease management. Future work will focus on validating the system across more diverse populations, including older adults and people with movement disorders, while reducing computational demands through model compression and edge computing to enable efficient, real-time operation on low-power wearable devices.
