- The value distribution curves of the reference patients or experiments
(fig.3_a) are used for the determination of reference percentiles (fig.3_b).
The numeric values of data columns are then converted for classification purposes
into triple matrix characters (0), (+), (-) depending of their
location between, above or below the calculated upper and lower percentile
threshold (see between fig.3_a/3_b). This is done for all numbers of a given
data column that is for reference (fig.3a) as well as for the abnormal
group data (fig.3_c)
- The operation results in triple matrix databases
suitable for data pattern classification with the CLASSIF1 algorithm.
- Symmetric percentile thresholds between 1 - 99%, 5 - 95%, 10 -90%,
15 - 85%, 20 - 80%, 25 - 75% or 30 - 70% are typically used to determine
the optimally discriminating percentile pair.
- Classifications can be further optimized by using either the optimal
percentile pair for all data columns or by combining the optimal percentile
pair of each data column.
