The generalization ability of minimizers of the empirical risk in the context of binary classification has been investigated under a wide variety of complexity assumptions for the collection of classifiers over which optimization is performed. In contrast, the vast majority of the works dedicated to this issue stipulate that the training dataset used to compute the empirical risk functional is composed of i.i.d. observations. Beyond the cases where training data are drawn uniformly without replacement among a large i.i.d. sample or modelled as a realization of a weakly dependent sequence of r.v.'s, statistical guarantees when the data used to train a classifier are drawn by means of a more general sampling/survey scheme and exhibit a complex dependence structure have not been documented yet. It is the main purpose of this paper to show that the theory of empirical risk minimization can be extended to situations where statistical learning is based on survey samples and knowledge of the related inclusion probabilities. Precisely, we prove that minimizing a weighted version of the empirical risk, refered to as the Horvitz-Thompson risk (HT risk), over a class of controlled complexity lead to a rate for the excess risk of the order $O_{\mathbb{P}}((κ_N (\log N)/n)^{1/2})$ with $κ_N=(n/N)/\min_{i\leq N}π_i$, when data are sampled by means of a rejective scheme of (deterministic) size $n$ within a statistical population of cardinality $N\geq n$, a generalization of basic {\it sampling without replacement} with unequal probability weights $π_i>0$. Extension to other sampling schemes are then established by a coupling argument. Beyond theoretical results, numerical experiments are displayed in order to show the relevance of HT risk minimization and that ignoring the sampling scheme used to generate the training dataset may completely jeopardize the learning procedure.