Personalized Weight Loss Management through Wearable Devices and Artificial Intelligence

TL;DR

Using wearable data from a 1-month trial involving around 100 subjects from the AI4FoodDB database, including biomarkers, vital signs, and behavioral data, key differences between those achieving weight loss (>= 2% of their initial weight) and those who do not are identified.

Abstract

Early detection of chronic and Non-Communicable Diseases (NCDs) is crucial for effective treatment during the initial stages. This study explores the application of wearable devices and Artificial Intelligence (AI) in order to predict weight loss changes in overweight and obese individuals. Using wearable data from a 1-month trial involving around 100 subjects from the AI4FoodDB database, including biomarkers, vital signs, and behavioral data, we identify key differences between those achieving weight loss (>= 2% of their initial weight) and those who do not. Feature selection techniques and classification algorithms reveal promising results, with the Gradient Boosting classifier achieving 84.44% Area Under the Curve (AUC). The integration of multiple data sources (e.g., vital signs, physical and sleep activity, etc.) enhances performance, suggesting the potential of wearable devices and AI in personalized healthcare.

Figures (5)

• Figure 1: Description of the application scenario and Machine Learning (ML) methods proposed in this study for the prediction of weight loss. The experimental framework integrates data acquired from Continous Glucose Monitor (CGM) devices and smartwatches. In particular, we consider the AI4FoodDB database romero2023ai4fooddb, considering five datasets covering physiological and lifestyle data, including Biomarkers (Dataset 4), Vital Signs (Dataset 6), Physical Activity (Dataset 7), Sleep Activity (Dataset 8), and Emotional State (Dataset 9). From these datasets, in the present article we propose the extraction of 284 total features that undergo feature selection before being evaluated using ML techniques. The proposed ML models ultimately distinguish between subjects who lost $\geq$ 2% of their initial weight and those who did not.
• Figure 2: Correlation matrix generated by the Pearson correlation coefficient. Positive correlations are marked in turquoise gradient, while negative correlations are marked in red gradient. The matrix reveals numerous strong correlations among features within the same dataset, indicating a linear relationship, but fewer correlations among features from different datasets.
• Figure 3: ROC curves for three different schemes. The first figure (top) shows the best configurations (feature selector and classifier) achieved in each scenario. The second figure (middle) includes the ROC curves for the Gradient Boosting models across different feature selectors for the combined datasets scenario. The third figure (bottom) shows the models for the combined datasets scenario using the Sequential Forward Floating Search feature selector.
• Figure 4: Comparative analysis of health metrics between participants who achieved significant weight loss ($>=2\%$) and those who did not ($<2\%$). Subject information and weight loss differences are also provided. Additionally, we include some specific feature information related to the best configuration achieved for DS4 (Biomarkers), DS7 (Physical Activity), and DS8 (Sleep Activity), described in detail in Table \ref{['tab:best_features']}.
• Figure 5: Comparative case study of two subjects: Subject A (35-year-old male, lost 5.4 kg) and Subject B (26-year-old female, gained 0.2 kg). General information (age, gender, height, etc.) and specific features are compared from the following datasets (DS): DS4 (Biomarkers), DS6 (Vital Signs), DS7 (Physical Activity), DS8 (Sleep Activity), and DS9 (Emotional State). Respon. Points = Responsiveness Points.