Publications on AI & Data Science

Peer-reviewed Journal Articles

Methodologies for Applying Artificial Intelligence to Physical Activity Interventions

In our comprehensive scoping review, we delve into the application of artificial intelligence (AI) in enhancing physical activity interventions. The review examines a range of AI methodologies – from machine learning to deep learning and reinforcement learning – in their role of advancing physical activity outcomes. The study analyzes 24 peer-reviewed articles, highlighting the effectiveness of AI in areas like pattern recognition, outcome prediction, and intervention improvement. It also identifies key future directions for AI in this field, such as personalized interventions and multimodal data integration. This pivotal work underscores the growing importance of AI in shaping effective physical activity strategies and promoting public health advancements.

Applications of Artificial Intelligence to Obesity Research

This scoping review provides a comprehensive overview of the applications of Artificial Intelligence (AI), specifically machine learning (ML) and deep learning (DL), in obesity research. The review, which analyzed 46 studies from PubMed and Web of Science, showcases how AI models have been effectively used to measure, predict, and treat obesity. It was found that AI models, in most cases, outperformed traditional statistical approaches in predicting obesity-related outcomes, indicating their potential in revealing complex patterns and relationships in obesity data. The review also observes a growing trend towards the use of advanced DL models for sophisticated tasks like computer vision and natural language processing. Furthermore, the review serves as an introductory guide to popular ML and DL models, highlighting their specific applications in the included studies. The study concludes by discussing future trends in AI applications in obesity research, such as multimodal models and synthetic data generation, underscoring AI’s evolving role in advancing obesity research methodologies.

Smart Sizing: AI’s Answer to Accurate Obesity Tracking

In this research, we developed machine learning (ML) models to enhance the accuracy of self-reported anthropometric data, a critical issue in monitoring population obesity risk. Utilizing data from 50,274 adults in the National Health and Nutrition Examination Survey (NHANES) from 1999 to 2020, the study addressed the substantial discrepancies often found between self-reported and objectively measured height, weight, and body mass index (BMI). The team applied nine ML models to predict these measures accurately. The most effective models significantly reduced the differences, decreasing the discrepancy in average height by 22.08%, weight by 2.02%, BMI by 11.14%, and obesity prevalence by 99.52%. Remarkably, the predicted obesity prevalence (36.05%) closely matched the objectively measured figure (36.03%), underscoring the potential of these models to provide reliable obesity prevalence estimates from survey data. This study demonstrates the promising application of ML in improving data accuracy for public health surveillance.

Nutrition at a Glance: AI’s Revolution in Nutritional Analysis of Nuts

In our study, we utilized deep neural network models to detect and quantify the nutritional content of common edible nuts from photographs. Our dataset comprised 1,380 images, each showcasing a variety of 11 popular nut types. Using transfer learning, the models were trained for precise multi-label classification and object detection. The model demonstrated a mean average precision of 0.7596 in nut localization and achieved a high accuracy of 97.9% in identifying nut types and quantities. Significantly, it accurately estimated the aggregate nutrient profiles, including total energy, protein, carbohydrates, fats, and essential vitamins and minerals, with an error margin between 0.8-2.6%. This advancement marks a significant step in incorporating AI into diet-tracking apps, offering potential benefits for nutritional tracking and promoting healthier dietary choices.

Pulse of the People: AI Decodes Social Media Sentiments on Policy

In two related studies, deep neural network models were employed to analyze public sentiment on Twitter regarding soda taxes and menu labeling regulations, key health policies aimed at combating obesity. The soda tax study, analyzing around 370,000 tweets from 2015 to 2022, observed a peak in public attention in 2016 with a subsequent shift towards neutral sentiment, while negative sentiments steadily increased over time. The menu labeling study, covering 2008 to 2022, also found that public discourse peaked around major policy announcements, with a trend towards neutral and news-related tweets, and a recent rise in negative sentiment. Both studies achieved high accuracy in sentiment classification and identified predictors like the author’s followers and tweeting frequency. These findings highlight the dynamic public attitudes toward health policies and underscore the potential of social media analysis in informing policy design and refinement.

Truth in Headlines: AI Tackles Misinformation in Obesity News

We developed artificial intelligence (AI) models to identify and correct news headlines that exaggerate obesity-related research findings, a critical issue that can mislead public perception and erode trust in scientific communication. The research involved collecting 523 exaggerated headlines from digital media, identifying common exaggeration patterns such as inferring causality from observational studies or generalizing findings inaccurately. The team created a BERT model fine-tuned to distinguish between exaggerated and non-exaggerated headlines, and developed generative language models (BART, PEGASUS, and T5) to automatically generate accurate headlines based on scientific abstracts. The performance of these models was high, with the BERT model achieving 92.5% accuracy and the generative models surpassing baseline ROUGE scores, demonstrating their potential to enhance the accuracy and integrity of media reporting on scientific research.

Chatbots and Body Image: Navigating the Digital Influence

In this study, we evaluated the responses of 14 popular chatbots to questions about body image, a crucial mental health concern, particularly among adolescents and young adults. The study focused on both companion and therapeutic chatbots, assessing their reactions to ten body image-related questions based on validated instruments. The findings revealed a modest overall quality in the chatbots’ responses, with scores averaging five out of nine and a wide range of individual scores from one to eight. Notably, companion chatbots tended to focus on comforting users, while therapeutic ones aimed more at identifying causes and suggesting remedies. Some therapeutic chatbots could recognize potential mental health crises. However, the study highlights substantial variability in the content and quality of responses among chatbots, raising concerns about the potential for misleading or biased advice. This underscores the need for continued technical and supervisory enhancements to ensure the safe and effective use of chatbots in sensitive areas like body image counseling.

Guarding Integrity: AI’s Role in Distinguishing Human vs Machine-Written Essays

In this study, a novel method is introduced to differentiate between machine-generated and human-written text, addressing the growing concerns about academic integrity and plagiarism in the context of advanced language models like ChatGPT. The study employed a dataset of student-written essays and comparable essays generated by ChatGPT, focusing on similarity scores within and between these sets. The methodology, which combines analysis of prompts and essays, demonstrated high accuracy in identifying machine-generated text, with impressive AUC, false positive, and false negative rates. This approach significantly outperforms traditional plagiarism detection tools that rely solely on text analysis. The research highlights the potential of this method in maintaining academic integrity but also acknowledges the need for further studies to assess its applicability in varied educational settings and with different model parameters.

Book

Using nearly a hundred tangible, real-world research examples, this guide presents a detailed pathway on how researchers, irrespective of their area of study, can employ ChatGPT as a competent research assistant. This guide encapsulates ten easy-to-follow principles to accomplish a vast array of research tasks:

Identifying research topics and framing questions through an in-depth discussion with ChatGPT
Formulating and refining hypotheses based on the chosen research question
Undertaking literature reviews, covering all steps of a systematic review protocol
Selecting adequate research design and corresponding methodology
Developing valid, reliable, and efficient research tools
Handling every aspect of data collection, management, and ethics
Interpreting and analyzing both quantitative and qualitative data
Writing and refining research papers and reports
Addressing peer review comments
Disseminating study findings through mass and social media platforms