Obesity disease is a significant health issue that has affected millions of people worldwide. A new study co-authored by Dr. Manjeet Rege, a professor in the School of Engineering at the University of St. Thomas, explores the application of ensemble learning techniques in predicting obesity risk using lifestyle data.
“By leveraging advanced algorithms in machine learning and artificial intelligence, it becomes feasible to build models capable of identifying individuals at risk of developing obesity,” said Rege, who is also the director of Center for Applied Artificial Intelligence at St. Thomas. “We hope this research contributes to the development of more effective obesity prevention and intervention strategies.”
The findings are in a new research paper titled, “An investigation of ensemble learning techniques for obesity risk prediction using lifestyle data,” which has been published in the Decision Analytics Journal, Elsevier.
In the health care sector, online medical repositories and hospitals are generating vast amounts of data, thereby providing valuable resources for researchers to explore and harness AI techniques to address real-life health issues, the research found.
“Identifying underlying reasons for the onset of obesity risk in its early stage has become challenging for medical practitioners,” Rege said. “The growing volume of lifestyle data related to obesity has made it imperative to employ effective machine learning algorithms that can gather insights from the underlying data trends and identify critical patient conditions.”
According to the research, Rege and his three co-authors – Dr. Shahid Mohammad Ganie, Dr. Hemachandran Kannan, and student Bobba Bharath Reddy, all of Woxsen University, India, – found that “the perfect climate for obesity to flourish has been created by traditional diets heavy in processed foods and low in physical exercise as technology develops and urbanization picks up speed. The resultant effects include an increase in disorders linked to obesity, such as diabetes, heart problems, and other ailments. Early detection of diseases and identification of associated risks can serve as crucial motivators for patients, encouraging them to make positive changes in their dietary habits, lifestyles, and exercise routines.”
Rege noted that early illness detection and the identification of risk factors can be extremely motivating for people. “With this knowledge, people are better equipped to start making healthy food, lifestyle, and exercise improvements,” he said. “It is especially beneficial to identify obesity-related concerns early on since it allows for timely treatment and lifestyle changes.”
Considering more than just BMI
Rege and the other researches noted that measurements from BMI, the body mass index, is typically what’s used as the primary indicator of obesity risk. But he noted that “although BMI is a useful measure, it does not capture the full complexity of obesity, which is influenced by a range of behavioral, environmental, and genetic factors.”
For example, they wrote that using BMI as the primary indicator of obesity does not account for muscle mass, fat distribution, or other important health indicators. And as a result, the findings might reduce the precision of obesity classification.
They also discovered that while some machine learning models for obesity classification may use lifestyle factors, sex and 3D body scans without relying on BMI, the dataset used could result in models that do not generalize well to new populations. That’s because some studies utilized smaller sample sizes from specific regions or countries, restricting the generalizability of the findings to larger or global populations.
For their research, Rege and his co-authors “developed a prediction model for obesity using boosting techniques but also conducted thorough comparisons among bagging, boosting, and voting models.” (Their paper defines the bagging, boosting, and voting ensemble learning techniques they used.)
“We selected three algorithms from each ensemble method, each with distinct characteristics and strengths, to demonstrate the effectiveness of our proposed model from multiple perspectives. We also employed preprocessing techniques to enhance the quality of the data,” Rege said.
Additionally, they used a publicly available dataset collected from the diverse demography of countries such as Colombia, Peru, and Mexico based on their eating habits, age, sex, and physical condition – they even considered factors such as water and alcohol consumption and the frequency of consumption of vegetables. This dataset includes a wide range of features, such as dietary patterns, physical activity levels, mental health, and sleep habits.
“We considered all 0–6 classes to distinguish between various obesity categories for more accurate disease prediction.”
They then applied those factors toward ensemble learning models that they developed, with that process described in the research paper.
They found that their approach contributes to a more comprehensive understanding of obesity risk factors, aiding health care providers in delivering targeted interventions based on specific obesity levels. They recognized, however, that a more balanced dataset would lead to a better prediction model. They stated that future research could explore the utilization of deep learning methods, potentially enhancing the detection and prediction of obesity risks.
