An Efficient Human Activity Recognition (HAR) Model Based on Convolutional Neural Networks for Computing Devices Aiming to Reduce Latency and Tackle the Inactivity of Gadgets

Human Activity Recognition (HAR) is examined using a High Resolution Convolutional Neural Network (HR-CNN) and is compared with Machine Learning (ML) baselines, such as Decision Tree (DT) and Support Vector Machine (SVM), and with Deep Learning (DL) models, including Artificial Neural Network (ANN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and a standard Convolutional Neural Network (CNN). Using a six-class dataset (10,299 samples), HR-CNN attains 99.2% accuracy and 0.88 macro-F1, outperforming all evaluated baselines. A supplementary analysis models latency as a regression target, and reports Mean Squared Error (MSE) and Coefficient of Determination (R²) across configurations. MSE reaches 0.008 compared with 0.031 in alternatives, with competitive R² values noted for both DT and CNN in distinct runs. Overall, HR-CNN delivers a state-of-the-art classification performance for HAR, while the regression study indicates favorable system-level behavior.

Ҳали таржима қилинмаган