The cellular users on-board operational trains heavily suffer from relatively poor performance of current deployment of cellular networks. The improvements in this context are very limited due to the fact that trains mainly operate on rural and slightly sub-urban environments characterized with numerous deep-fade events occurring frequently. Another major concern comes from quasi-metallic train structure, where vehicle penetration loss (VPL) becomes severe and thus seriously affects voice and data applications. However, this impact can be significantly minimized by equipping the train with amplify-and-forward (AF) relay with attached radiating cable. Therefore, to benchmark and understand the variations due to various propagation conditions we conduct real-world measurements on operational trains. In this paper, we address the issue of identifying nominal environments based on statistics of coverage metrics by using agglomerative information bottleneck (AIB). Based on partitions of AIB, we formulate the problem of change detection with nonparametric hypotheses and show that this approach significantly matches to regressions obtained from special case of logistic function (LF) approach.