Human activity recognition is the process of identifying a person’s physical movements. A one-time series classification technique that is useful in many different topics, contexts, and applications is Human Activity Recognition (HAR). Regular activities, including walking, jogging, standing, and sitting, are frequently performed. Human Activity Recognition tasks are often handled by deep learning architectures with clear results. Using the Long Short-Term Memory-Gated Recurrent Unit (LSTM-GRU) and LSTM-Bidirectional GRU (BiGRU) architectures, movements or activities are performed using the University of California, Irvine (UCI) HAR dataset. The proposed models completed the training, optimization, and accuracy testing. The accuracy score was 88.02%. The losses, accuracy, training iterations, and other details are displayed using a graph and confusion matrix.