Estimation of feature importance for food intake detection based on Random Forests classification

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Pattern Recognition

Background:

• Accurate food intake detection is crucial for health monitoring and dietary studies.
• Wearable sensor systems offer a promising approach for unobtrusive, continuous monitoring.
• Feature selection is critical for optimizing the performance of pattern recognition systems.

Purpose of the Study:

• To investigate the importance of time domain (TD) and frequency domain (FD) features for automatic food intake detection.
• To evaluate feature relevance using Random Forests classification within a wearable sensor system.
• To identify the most representative features for distinguishing food intake events.

Main Methods:

• Utilized the Automatic Ingestion Monitor (AIM) with jaw motion, hand gesture, and accelerometer sensors.
• Collected 24-hour free-living data from 12 subjects with unrestricted food intake.
• Extracted TD and FD features from sensor signals and employed Random Forests for feature importance estimation.

Main Results:

• Frequency domain (FD) features from the jaw motion sensor were highly relevant.
• Time domain (TD) features from the accelerometer signal also proved significant for detection.
• The Random Forests classifier effectively estimated the importance of various extracted features.

Conclusions:

• Specific TD and FD features from particular sensor modalities are most effective for automatic food intake detection.
• Jaw motion (FD) and accelerometer (TD) data provide the most discriminative information for ingestive behavior monitoring.
• This research informs the design of more accurate and efficient wearable systems for dietary assessment.