Efficient Human Activity Recognition through Multi-Sensor Data and Deep Learning Techniques

Extensive research takes place in the field of Human Activity Recognition (HAR) systems with wearable computing that helps in healthcare applications, fitness tracking and smart environments. Sensor technology advances have propelled Human Activity Recognition systems to utilize innovative machine learning methods for accurate and robust activity classification. This work evaluates the effectiveness of four cutting-edge deep learning models-Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN) and Gated Recurrent Units (GRU)-through their implementation on publicly accessible accelerometer and gyroscope HAR datasets. These classification models underwent testing through different activation functions namely ReLU, Leaky ReLU, and Swish to assess their practicality for classification operations. Standard evaluation metrics including accuracy, precision, recall, ROC-AUC and F1-score demonstrate the best features of each implemented model while providing descriptive visual representations. TCN delivered 98.59% accuracy and exhibited superior training speed efficiency thus establishing itself as the suitable AI algorithm for real-time implementation. This research offers an extensive breakdown for choosing optimal models in real-time HAR systems which supports advances in wearable computing and activity recognition technologies.