5. Results
The autocorrelation assessment results revealed that all the fall
indicator time series were autocorrelated. After ARIMA modeling,
non-autocorrelated fitted residuals were obtained for all the fall
indicator time series.Fig. 3illustrated the results of one of the fall
indicators from the autocorrelation assessment and ARIMA modeling for one participant. In particular, Fig. 3a showed the original
time series.Fig. 3b showed that the ACF plot did not cutoff at Lag
10 indicating that this fall indicator time series were autocorrelated. Differenced time series were then determined (e.g.Fig. 3c). It
was found that the ACF plot (Fig. 3d) and PACF plot (Fig. 3e) for
the first order differenced time series had ‘‘tail off’’ pattern and
cut off pattern after Lag 2, respectively. This indicated that an ARIMA (2, 1, 0) should be appropriate. A residual time series (Fig. 3f)
was then obtained. The diagnosis check revealed that the residual
time series were not autocorrelated, since the autocorrelation coefficient decay rapidly to the value that was not significantly larger
than zero at Lag 10 (Fig. 3g)