In this seminar spline quasi-interpolation is presented as a valid tool to enhance the performance of unsupervised-learning based approaches in the context of clustering. Due to its local nature, quasi-interpolation has a reduced computational cost compared to global methods (like interpolation); moreover, data collected from realistic scenarios are usually affected by errors, therefore, global methods might be too stringent, beside suffering from overfitting. As a first application, quasi-interpolation is adopted to derive a smoothing model that can be combined with common clustering techniques for the task of anomaly detection in time-series data. Secondly, in the context of model based clustering, quasi-interpolation can be used to construct an estimation of the monovariate empirical marginal densities which underlie suitable multivariate distributions of the given data. The adopted quasi-interpolant operator is proved to be a uniformly consistent estimator of the sought density which outperforms classical approaches. In particular, copulas will be employed to construct the final clustering model. For both the applications, numerical tests will be shown on artificial and real datasets.