Mitigating underreported error in food frequency questionnaire data using a supervised machine learning method and...

Area of Science:

• Nutritional Epidemiology
• Biostatistics
• Machine Learning in Health

Background:

• Food frequency questionnaires (FFQs) are vital for diet-disease research but suffer from self-reporting biases like social desirability and misclassification.
• Existing methods attempt to model and correct measurement error in dietary data.
• Addressing these biases is crucial for reliable nutritional epidemiology.

Purpose of the Study:

• To propose a novel machine learning method to adjust for measurement error in FFQ data.
• To specifically address and correct underreporting in dietary intake.
• To enhance the reliability of FFQ data for epidemiological studies.

Main Methods:

• A random forest (RF) classifier was employed to label FFQ responses.
• A custom algorithm was developed to adjust for measurement error based on RF classifications.
• The method was validated using both participant-collected and simulated FFQ data.

Main Results:

• The proposed method achieved high model accuracies, ranging from 78% to 92% for participant data.
• An accuracy of 88% was obtained for simulated data.
• The results demonstrate the efficiency of the RF classifier and error adjustment algorithm in correcting underreported dietary entries.

Conclusions:

• The novel machine learning approach effectively corrects underreported data in FFQs.
• This method offers a valuable tool for nutrition researchers to improve dietary data quality.
• The technique can be used independently of diet-disease models, reducing noise and enhancing subsequent analyses.