A new procedure for combined validation of learning models – developed for specifically uncertain data – is briefly described; it relies on a combination of resubstitution with the modified learn-and-test paradigm, called by us the queue validation. In the initial experiment the elaborated procedure was checked on doubtful (presumably distorted by creative accounting) data, related to small and medium enterprises of the Podkarpackie-region in Poland. Validated in the research learning models were completed in the form of decision trees and sets of production rules. Correctness of both type of models (trees and rules) was estimated basing on the error rate of classification. It was found that false-positive classification errors were significantly larger than false-negative ones; the difference discovered by validation procedure can be probably used as a hint of fraud in the evaluated data.
Citation rules
Licence
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.