A Transformer-Based Hybrid Model for Human Activity Recognition Using Smart Home Environmental Sensor Data

With the rapid expansion of smart homes, accurate and automatic recognition of human activities has become one of the key challenges in the fields of artificial intelligence and the Internet of Things. This technology has vital applications in areas such as elderly care, health monitoring, and enhancing the security of smart homes. In this research, a deep learning-based hybrid approach for human activity recognition is introduced, which utilizes transformer models and gated recurrent units. The transformer model, with its multi-head attention mechanism, has the ability to analyze long-term relationships among sensor data and identify behavioral patterns with higher precision. The gated recurrent unit, due to its capability in learning temporal patterns, significantly contributes to improving the accuracy of activity recognition.Evaluation results show that the model achieved an accuracy of 95.19% on the Aruba dataset and 89.01% on the Milan dataset, indicating the high generalizability and pattern recognition ability of the proposed model. Furthermore, compared to similar methods, the proposed model has demonstrated a remarkable performance in improving human activity recognition.