Human Activity Recognition with smartphone sensors data using CNN-LSTM

Human activity recognition (HAR) utilizing wearable inertial sensors has nowadays become a new research hotspot due to various advancements in sensor technology. Recently, methods based on deep learning (DL) have been successfully used to evaluate time series data recorded by wearable sensors and smartphone to predict a variety of human activity. Furthermore, the majority of HAR methods relied on traditional feature engineering. In this study, a convolutional neural network along with, a long short-term memory network, that is deep learning architecture for activity recognition is proposed here (CNN-LSTM). With minimum data pre-processing, the proposed model CNN-LSTM network extracts feature from raw sensor data automatically. By using testing data, the trained model is then used for the recognition of various activity. The performance of the model is examined using the benchmark dataset WISDM. The results of the experiment show how the suggested approach outperforms alternative strategies and how adaptable it is to be implemented in sensor-based healthcare systems for recognition of human activity. The research findings show the proposed model outperforms the other approaches that were compared by achieving 98.04% accuracy and a F1 score of 98.04%, which is better than the similar existing models.