A comparison of strategies for imputing saturated pressure array data with application to the wheelchair-seating interface.

PURPOSE: The common responses to pressure sensor saturation are extreme: either discarding of data, or wholesale alteration of experimental protocol. Here, we test four simplistic strategies for restoring missing data due to sensor saturation, avoiding such drastic measures. METHODS: We tested these algorithms on 62 pressure maps collected from 42 individuals (20 M/22 F, 54.1 ± 26.2 years, 1.7 ± 0.1 m, 71.9 ± 17.8 kg) under a variety of seating conditions. These strategies were tested via a cross-validation design, censoring the maximum pressure value in the datasets and measuring prediction error. RESULTS: The four strategies showed various prediction error rates: ? = 0.43 ± 0.14 (simple substitution), ? = 0.16 ± 0.21 (scaled substitution), ? = 0.19 ± 0.21 (feature extraction), and ? = 0.24 ± 0.32 (extrapolation by non-linear modeling). CONCLUSION: For single-sensor saturation, it may be possible to restore missing data using simple techniques. Implications for Rehabilitation We present a method for imputing missing data from pressure sensor arrays. The implications for rehabilitation are as follows. Improved flexibility in design of protocols concerning interfacial pressure measurement. Restoration of missing data from existing datasets. Reduction in recruitment burden for future studies. Reduction in exposure risk to study participants.