Mobile Ad-hoc Networks are used in many scenarios (e.g., emergency management) for supporting collaborative work of operators. But this requires either (i) continuous connections, or at least (ii) the possibility to foresee that a device is going out to disconnect (e.g., in order to locally cache important data needed for the following activities, etc.). Therefore a basic problem is how to predict possible disconnections of devices, in order to let upper layers ap- propriately address connection anomalies (e.g., either tak- ing global remedial actions to maintain the network con- nected, or local ones to let the disconnecting device to go on for some time with its own work). In this paper we present a bayesian approach to predict disconnections in MANETs, and validating experimental results that shows the viability of the approach.