Abstract
Many statistical challenges arose in the advertisement field, such as displaying an advertisement at the right place and the right time. To achieve this, several performance metrics of airing of an ad were measured over time, for example the number or rate of clicks on an advertisement. A crucial aim is to be able to predict the clicks rate. Nevertheless, it is essential to understand beforehand the voluminous and heterogeneous data structure. Therefore, we introduce a clustering method of publicity campaign tim-series, i.e. of non-Gaussian longitudinal data, with a mixture of generalized linear models (GLM).