Various context mining algorithms are proposed aiming at improving accuracy. The major problem of the existing algorithms is that they assume no data loss in sensor data input, thus are unable to function sufficiently in a practical environment, where sensor data frequently drop. This paper proposes a novel middleware system called Uninterruptible Data Supply (UDS) system, which compensates the missing data with a probabilistic manner. Applications can benefit from UDS on the occurrence of constant and temporal deficit of sequential or discrete data. The evaluation shows that UDS can sustain the accuracy of context over 80% even with 40% data missing.