To that data processing

The data-processing audit (also called audit of the Information systems) is the evaluation of the risks of the computer operations, with an aim of bringing a reduction in those and of improving the control of the information systems.

The listener bases himself on the reference frames according to:

COBIT (describing the complete operation of a data processing department)
Standard ISO

cf isaca or afai

---- the following part does not correspond to the definition of the data-processing audit retained by the profession: ----

There exist two types of to that data processing :

the audit of the need;
the audit of discovered knowledge.

Audit of the Need

The audit of the need comprises two parts:

analysis of what exists;
determination of the target.

The analysis of existing consists of a work of ground at the end which one formalizes the circulation of the documents " types" from one actor to another and the treatment which each actor applies to these documents, using Logigramme S (treatments on the documents) and of representation of relational Graphe S (circulation of the documents).

The determination of the target consists in locating:

the passages " paper - numérique" and " numerical - papier" ;
redundancies in the graph (" forms in several exemplaires") ;
bottlenecks (dispersion of the infrastructures, check-points and validation necessary?) ;
cutting in zones, subgraphs: the " fields; métier". Thus each zone can be equipped with a specific tool, rather than only one total system " machine with gaz".

Thus the result of the audit, generally worked out and collectively approved, can give place not to a project, but several projects scheduled in a roadmap.

Audit of Discovered Knowledge

The audit of discovered knowledge consists in developing the existing data and knowledge in the company. The mathematical modeling of the databases always obliges with " perdre" part of information, it acts to redécouvrir it in " brassant" data.

One audit of discovered knowledge generally leads to the assembly of a decisional system, but also to be the prelude to a management system of knowledge.

The audit of discovered knowledge is practiced in the following way:

No objectives: one does not know what one will discover;
a field: one knows about which trades one works, therefore about which databases;
an integrated team: work in situ, three actors of which an expert " métier" , an expert " administration informatique" and a digger of data (see Dated mining)
work is done by short loop of prototyping
to facilitate the mixing of the data, one modifies their format and their relative tendency. It is the " preprocessing" who takes the most time
Using algorithms with training on the one hand, and of visualizations of data on the other hand, the digger highlights empirical bonds between the data
the correlations and bonds selected must answer three criteria: unknown to the user, explainable a posteriori and useful. The theoretical demonstration of the phenomenon is completely superfluous.
detected knowledge is formalized (trees, graphs, tables, rules, etc) then prototypées logiciellement
the validity of prototypées knowledge is checked thanks to a statistical test (khi two, kappa, etc) on a data file known as " test set" , different from the play having been used for the analysis (" training set")
the test set can be either external with training set, or recomputed from him (rééchantillonnage)
If the test passed, the model is put in production
One starts again the cycle.

Concretely, the stacking of the models, themselves sometimes complex (trees and Récursivité) can lead to truths problems of data-processing architecture:

Parallelization of calculations
Volumetry of the data
Load of the network, partly due to the mechanisms of replication

On another side, the computation result resulting from a stacking of models can lead to results particularly relevant and surprising.

Random links: