Hybrid CNN-LSTM Model for Accurate Human Activity Recognition Using Wearable Sensor Data

Abstract

Human Activity Recognition (HAR) is an important activity in artificial intelligence, machine learning that can be applied in health, surveillance, fitness monitoring, and intelligent environments. In this paper, the researcher suggests using a hybrid CNN-LSTM deep learning model to identify human activities based on wearable sensor data properly. The training model is trained on cleaned and pre-processed PAMAP2 data. activities under consideration include lying, sitting, standing, walking, cycling, and vacuum-cleaning. Convolutional layers are used to effectively extract spatial features, whereas the LSTM layers are used to detect temporal relationships in sensor data. The architecture suggested already has a high overall accuracy of 98%, as well as other significant results (precision, recall, and F1-score). These findings prove that the model can be used in real-time applications because it has a lightweight representation and good handling of diverse activity classes. The paper also points out some of the major problems in HAR and provides potential solutions by noting attention mechanisms, using multimodal sensors, and deploying on edge devices.