ActiNet: A reliable Deep Learning model for Human Activity Recognition using Internet of Things

Utilizing sensor data to recognize human activity is a difficult and significant undertaking for wearable technologies and health monitoring. This work presents a new deep learning model, named ActiNET, which demonstrates exceptional accuracy and resilience in Human Activity Recognition (HAR). The preprocessing of raw sensor data from accelerometers, gyroscopes, and pulse rate monitors in our model involves Robust Scaling techniques. Next, it utilizes an ActiNET architecture that integrates Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks to extract both spatial and temporal information from the input. The research also employed the Synthetic Minority Over-sampling Technique (SMOTE) to equalize the dataset and enhance the performance of the model. The model is assessed using a dataset that we have carefully selected ourselves, which encompasses five distinct human activities: walking, standing, sitting, running, and sleeping. Our model surpasses conventional HAR algorithms, with an accuracy rate of 94.88% as demonstrated by the testing findings. The study showcases the capabilities of Internet of Things (IoT) and deep learning in HAR, and offers valuable insights for further investigation in this domain.