Factors influencing the reliability of statistical body surface potential map classification

The maximal distance between samples taken from the same multivariate distributions was studied experimentally as a function of the sample size and dimensionality using body surface potential map (BSPM) data from normal subjects. Experimental findings show that for a given sample size, the involvement of too many parameters (dimensionality) results in an unreliable conditional cumulative probability density (CCPD) estimate. Consequently, the diagnostic performance during classification can be significantly lower than anticipated on the basis of an inadequate training set. For a given sample size, there is an optimal dimensionality, below which representation accuracy and above which the CCPD estimation reliability are heavily compromised.<
>