A hybrid multi-layer perceptron with selective stacked ensemble learning approach for recognizing human activity...

Area of Science:

• Computer Science
• Machine Learning
• Signal Processing

Background:

• Human Activity Recognition (HAR) is crucial for monitoring human movement using sensor data.
• Existing HAR methods face challenges in accuracy and efficiency, particularly in cross-domain applications.
• Developing robust and efficient HAR systems is vital for real-world IoT applications.

Purpose of the Study:

• To introduce a novel Hybrid Multi-Layer Perceptron (MLP) model for enhanced Human Activity Recognition.
• To improve accuracy and computing efficiency in HAR, especially for cross-domain generalization.
• To evaluate the model's performance against conventional methods and establish its suitability for real-world systems.

Main Methods:

• A Hybrid MLP model was developed, integrating Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Logistic Regression (LR) in a stacked ensemble.
• Smartphone accelerometer and gyroscope data were fused and processed through the MLP for feature extraction.
• The model was trained on a source HAR dataset and evaluated on the PAMAP2 dataset for cross-domain performance.

Main Results:

• The proposed Hybrid MLP model achieved up to 99% accuracy on the source HAR dataset, outperforming KNN, DT, MLP, and CNN.
• In cross-dataset evaluation, the model outperformed a CNN baseline by 2% in accuracy and F1-score on the PAMAP2 dataset.
• The model demonstrated competitive performance compared to the CrossHAR method without requiring computationally intensive pretraining.

Conclusions:

• The Hybrid MLP model offers high accuracy and efficient inference for Human Activity Recognition.
• The model exhibits strong cross-domain adaptability, making it suitable for diverse real-world scenarios.
• This approach provides a practical solution for developing efficient and accurate IoT-based activity recognition systems.