Markov chain-based models for missing and faulty data in MICA2 sensor motes

We have developed Markov chain-based techniques for infield modeling the missing and faulty data for the widely used MICA2 sensor motes. These models help designers of sensor nodes and sensor networks to gain insights into the behavior of any particular sensor platform. The models also enable users of sensor networks to collect high integrity data from the deployed networks in a more efficient and reliable way. The new approach for development and validation of faults and missing data has two phases. In the first phase, we conduct exploratory analysis of data traces collected from the deployed sensor networks. In the second phase, we use the density estimation-based procedure to derive semi Markov models that best capture the patterns and statistics of missing and faulty data in the analyzed sensor data streams. We have applied the fault detection and missing data modeling procedure on light, temperature and humidity sensors on MICA2 motes in sensor networks deployed in office space and natural habitats. The technical highlight of the research presented in this paper include: (i) exploratory data analysis and studying the properties of the sensor data streams; and (ii) adoption of a new class of semi Markov-chain models for capturing and predicting missing and faulty data in actual data trace streams