Comparison of different classifiers with varying amounts of imprecise data.

This table compares the performance of different classifiers for the original MIMIC II data and a version of the MIMIC II data where 50% of observations were randomly perturbed by a value ϵ, distributed normally with mean zero and the empirical variance of the variable in question. We see that N-UFA is robust to imprecise data, with accuracy decreasing 1.7% as the amount of imprecise data increases to 50%. An expanded version of Table 7 including confidence intervals and results for 5–25% imprecise data is available in the S2 Table.