Enhancing human activity recognition with machine learning: insights from smartphone accelerometer and magnetometer...

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human Activity Recognition (HAR) has advanced significantly due to sensor and AI progress.
• Previous studies established baseline results for HAR using smartphone sensor data.

Purpose of the Study:

• To implement and evaluate various machine learning techniques for improved HAR.
• To analyze the effectiveness of neural networks, random forest, and other models on existing HAR datasets.

Main Methods:

• Data collection from smartphone sensors, followed by cleaning and normalization.
• Feature extraction and implementation of diverse machine learning models including neural networks and random forest.
• Utilized non-normalized data and integrated magnetometer signals for enhanced performance.

Main Results:

• Neural network and random forest models demonstrated high effectiveness.
• Achieved an Area Under the Curve (AUC) of 98.42%, classification accuracy of 90.14%, F1-score of 90.13%, precision of 90.18%, and recall of 90.14%.
• Outperformed earlier models with reduced computational cost, comparable to deep learning approaches.

Conclusions:

• The developed approach is novel, efficient, and suitable for real-time mobile applications.
• Lightweight models and a reproducible visual workflow enhance deployability.
• The integration of non-normalized data and magnetometer signals improved HAR performance.