Multi-Sense Attention Network (MSANet): Enhanced Human Activity Recognition Using Deep Learning Architectures with Self-Attention Mechanisms

Multi-Sense Attention Network (MSANet), a sophisticated deep learning framework designed for Human Activity Recognition (HAR) from wearable sensor data and we are introducing it in the paper.MSANet integrates convolutional neural networks (CNNs), recurrent neural networks (RNNs), and self-attention mechanisms to exploit both spatial and temporal features effectively.The architecture is uniquely structured with multi-filter convolutional layers, identity mappings, and convolutional skip connections that enrich feature extraction and processing capabilities.A key innovation in MSANet is the implementation of self-attention layers, which enhance the model's focus on pertinent features critical for accurate activity classification.Evaluated on the UCI HAR dataset, MSANet achieved an impressive overall accuracy of 97.62% on the test set, with robust precision, recall, and F1-scores across different activities.These results demonstrate MSANet's potential in advancing the precision of HAR systems, making it a valuable tool for applications in healthcare, fitness tracking, and beyond. CCS Concepts Computing methodologies Activity recognition and understanding.