Guided regularized random forest feature selection for smartphone based human activity recognition.

Area of Science:

• * Human Activity Recognition (HAR)
• * Sensor Data Analysis
• * Machine Learning for Health

Background:

• * Human Activity Recognition (HAR) is crucial for applications like remote health monitoring and smart homes.
• * Smartphone sensors generate high-dimensional data, necessitating efficient feature selection.
• * Current methods often struggle to balance feature relevance, redundancy, and classification accuracy.

Purpose of the Study:

• * To develop an efficient HAR model using smartphone sensor data (accelerometer and gyroscope).
• * To extract relevant time-frequency domain features without data loss.
• * To enhance HAR classification accuracy and reduce recognition time.

Main Methods:

• * Proposed a feature selection method using Guided Regularized Random Forest (GRRF).
• * Utilized time-frequency domain features from 3-axial accelerometer and gyroscope data.
• * Employed a Support Vector Machine (SVM) for activity classification post-feature selection.

Main Results:

• * Achieved high classification accuracies: 99.10% on a self-collected dataset and 99.30% on the UCI HAR dataset.
• * Demonstrated the effectiveness of the GRRF-based feature selection in identifying pertinent and non-redundant features.
• * Validated the model's generalization capability on multiple datasets.

Conclusions:

• * The proposed GRRF feature selection method significantly improves HAR efficiency and accuracy.
• * Time-frequency domain features combined with GRRF and SVM offer a robust approach for HAR.
• * This efficient HAR model has strong potential for real-world applications in health and smart environments.