Ensemble of One-Class Classifiers for Personal Risk Detection Based on Wearable Sensor Data.

Area of Science:

• Machine Learning
• Biomedical Engineering
• Data Science

Background:

• Ensembles of one-class classifiers are used in various applications.
• Existing methods lack focus on personal risk detection.
• Wearable sensors generate data for user monitoring.

Purpose of the Study:

• Introduce the One-Class K-means with Randomly-projected features Algorithm (OCKRA).
• Improve personal risk detection performance using the PRIDE dataset.
• Evaluate OCKRA against established classification algorithms.

Main Methods:

• OCKRA is an ensemble of one-class classifiers.
• It utilizes multiple random feature subset projections.
• The Personal RIsk DEtection (PRIDE) dataset, from 23 subjects using wearable sensors, was used for evaluation.

Main Results:

• OCKRA outperformed Support Vector Machine and Parzen window classifiers.
• Average Area Under the Curve (AUC) improvement was at least 0.53%.
• OCKRA achieved over 90% AUC for more than 57% of users.

Conclusions:

• OCKRA demonstrates superior performance in personal risk detection.
• The algorithm shows promise for applications utilizing wearable sensor data.
• OCKRA offers a significant advancement in detecting personal risk.