The Performance of an Algorithm for Classifying Gym-based Tasks across Individuals with Different Body Mass Index

ABSTRACT Previous activity classification studies have typically been performed on normal weight individuals. Therefore, it is unclear whether a generic classification algorithm could be developed that would perform consistently across individuals who fall within different BMI categories. Acceleration data were collected from the hip and ankle joints of 50 individuals: 17 normal weight, 14 overweight, and 19 obese. Each participant performed a set of 10 dynamic tasks, which included activities of daily living and gym-based exercises. The performance of a generic classification algorithm, developed using linear discriminant analysis, was compared across the three separate BMI groups for each sensor. Higher classification accuracies (92–95%) were observed for the ankle sensor; however, both sensors demonstrated consistent performance across the three groups. This is the first study to demonstrate the effectiveness of a generic classification algorithm across individuals with different BMI and may be a first step toward automated activity profiling in weight-loss programs.