Statistical Analysis Of Window Sizes And Sampling Rates In Human Activity Recognition

Accelerometers are the most common device for data collection in the field of Human Activity Recognition (HAR). This data is recorded at a particular sampling rate and then usually separated into time windows before classification takes place. Though the sampling rate and window size can have a significant impact on the accuracy of the trained classifier, there has been relatively little research on their role in activity recognition. This paper presents a statistical analysis on the effect the sampling rate and window sizes on HAR data classification. The raw data used in the analysis was collected from a hip-worn Actigraphy G3X+ at 100Hz from 77 subjects performing 23 different activities. It was then re-sampled and divided into windows of varying sizes and trained using a single data classifier. A weighted least squares linear regression model was developed and two-way factorial ANOVA was used to analyze the effects of sampling rate and window size for different activity types and demographic categories. Based upon this analysis, we find that 10-second windows recorded at 50Hz perform statistically better than other combinations of window size and sampling rate.