Impact of Reduced Sampling Rate on Accelerometer-Based Physical Activity Monitoring and Machine Learning Activity Classification

Purpose : Lowering the sampling rate of accelerometers in physical activity research can dramatically increase study monitoring periods through longer battery life; however, effect reduced on metric validity is poorly documented. We therefore aimed to assess measuring both overall and by specific behavior types. Methods Healthy adults wore sets two Axivity AX3 dominant wrist hip for 24 hr. At each location one accelerometer recorded at 25 Hz other 100 Hz. Overall acceleration magnitude, time moderate vigorous activity, behavioral activities were calculated processed using linear nearest neighbor resampling. Correlation between magnitude classifications rates was regression performed. Results Of 54 total participants, 45 contributed >20 hr wear 51 time. Strong correlation observed 25- 100-Hz measurement ( r = .97–.99), yet consistently lower data collected (3.1%–13.9%). Reduced sleep light increased sedentary classified 25-Hz machine learning models. Discrepancies greater when interpolation resampling used postprocessing. Conclusions The are highly correlated with predictable differences, which be accounted interstudy comparisons. Sampling methods should reported studies, carefully considered design, tailored outcome interest.