Improving aircraft passenger (dis-)comfort prediction via time series analysis of seat pressure data.

This study investigates whether passengers’ perceptions about aircraft seat (dis-)comfort can be predicted through Time Series Classification (TSC) of seating pressure indexes. Specifically, the study compares the performances of several TSC algorithms applied in a univariate as well as a multivariate setting to predict seat discomfort through some well-known seat pressure indexes. Results reveal that the lowest predictive performances are obtained when predicting seating discomfort using the time series of the peak pressure index, whereas the highest performances are generally obtained when using the SPD (Seat Pressure Distribution) index, which measures the ability of the seat to evenly distribute body pressure. The study also investigates the impact of data augmentation (DA) on the performance of the TSC algorithms in both univariate and multivariate setting; results reveal that the use of DA produces a satisfactory improvement in predictive performance of ResNet algorithm in multivariate TSC.