Optimizing Real-Time Action Recognition: A Hybrid CNN-LSTM Approach with Advanced Activation and Optimization Techniques

Wearable technology has significantly increased the importance of human activity recognition (HAR), which tracks physical activities for healthcare, fitness, and computer interaction. A new model combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks has been developed to classify different physical activities more accurately. The GELU activation function and Adamax optimizer achieved the best classification accuracy at 98.3%, outperforming other combinations. The combined CNN-LSTM model is about 98% accurate in distinguishing different activities, highlighting the importance of activation functions and optimizers in HAR applications.