A suboptimal state estimation algorithm is proposed for a discrete time stochastic multi-mode switched system, with a finite number of modes. The system state is transferred from one mode to the next. The parameters in the different modes are assumed to be known, but the mode-switching is random. The proposed filter is derived by extending the innovations approach to the problem of estimating the state of the system assuming that the switch between modes is purely random, but with known mode probabilities. As a special case we derive the filter for a linear system degraded by intermittent sensor failure and analyze its performance. This model is an alternative to the packet loss model, where the presence or absence of a packet is part of the observed data.