Are Machine Learning Algorithms More Accurate in Predicting Vegetable and Fruit Consumption Than Traditional...

Area of Science:

• Nutrition science
• Data science
• Behavioral science

Background:

• Dietary choices and behaviors are influenced by complex interactions.
• Machine learning (ML) offers potential for understanding these complex relationships.
• Predicting adequate vegetable and fruit (VF) consumption is crucial for public health.

Purpose of the Study:

• To compare the accuracy of ML algorithms against traditional statistical models in predicting adequate VF consumption.
• To explore the predictive performance of various ML algorithms using a comprehensive feature set.

Main Methods:

• Utilized a large dataset (2,452 features from 525 variables) from 1,147 French-speaking adults.
• Defined adequate VF consumption as 5 servings/day, measured via web-based 24-hour recalls.
• Compared nine ML algorithms (including SVM) with logistic regression and Lasso, performing data normalization and sensitivity analyses.

Main Results:

• Logistic regression and Lasso achieved an accuracy of 0.64 in predicting adequate VF consumption.
• Support Vector Machine (SVM) models with radial basis or sigmoid kernels showed the highest accuracy at 0.65.
• SVM with a linear kernel was the least accurate ML algorithm (0.55 accuracy).

Conclusions:

• ML algorithms and traditional statistical models demonstrated comparable accuracy in predicting adequate VF consumption in adults.
• The study suggests that ML's potential in predicting complex dietary behaviors requires further investigation.
• Additional research is necessary to fully leverage ML for understanding multifaceted dietary behaviors.