Quality characterization of a Long Term Evolution (LTE) cellular network with Multiple Input Multiple Output (MIMO) configuration is carried out through an experimental multivariate analysis of the main parameters of signal quality, which is crucial to optimize network performance. We adopted a technique based on the Expectation-Maximization (EM) algorithm that aims at statistically model radio-layer parameters with a blind machine learning technique that clusters data collected by a mobile operator. Data are retrieved with a smartphone-based methodology during a drive-test campaign. Clustering of the performance indicators has also been done spatially, by locating areas with different levels of signal quality on a map, to highlight those spots were improvements are required to overcome porr signal quality mostly due to the presence of co-channel or adjacent channel interference.