This thesis deals with two main algorithmical domains: Data Mining and Arithmetical computations. In both, we are interested in the average-case analysis of algorithms, and, we adopt more precisely the dynamical analysis point of vue which is a mixed method between Analysis of Algorithms and Dynamical Systems. The Euclid algorithms compute the gcd of two numbers; these are fundamental blocks in computer algebra, but their fine probabilistic behavior is always unknown. Thanks to Dynamical Analysis methods, recent important results have been obtained. In this thesis, we extend this approach to a precise analysis of parameters, as the binary complexity or the size of remainders. These parameters are essential for the Divide and Conquer gcd algorithm due to Knuth-Schönhage. Dynamical Analysis is also used for proven computations of spectral constants. The dynamical approach is then adapted to on polynomial Euclid algorithms even if, in this case, classical Analytic Combinatorics already applies. We also deal with Data Mining. We restrict ourselves to binary databases where the knowledge is represented by 'frequent patterns'. The number of frequent patterns is essential for analysing algorithms but experiments show that it significantly changes with the parameters of the database. Then, the worst case analysis is not meaningful in practice. In this thesis, we elucidate the average beahvior of the number of frequent patterns under a large model of databases built with eventually correlated sources
