Modeling unreliable data and sensors: Analyzing false alarm ratesusing training weighted ranking
Conference
Iyer, V, Iyengar, SS, Murthy, GR et al. (2011). Modeling unreliable data and sensors: Analyzing false alarm ratesusing training weighted ranking
. 171-176.
Iyer, V, Iyengar, SS, Murthy, GR et al. (2011). Modeling unreliable data and sensors: Analyzing false alarm ratesusing training weighted ranking
. 171-176.
Distributed Sensing allows live data retrieval and information processing of many parameters, which monitor events and phenomenon of interest. Live data retrieval uses spatial correlation and pre-processing to minimize data redundancy, while datamining helps build a statistical model for the aggregated sensor data. Managing the huge data collection and its data retrieval needs good understanding of the user's preferences, which otherwise would be over fitting and biased. This is not achievable without prior ranking of the datasets. In this paper, we work with two forms of ranking weights, precession and relevance, which helps in reliably finding hidden patterns and predicting future event. We evaluate the ranking function with UCI Machine learning data repository datasets and show that the relationship of precession and relevance holds good, and influence of aggregate network data allows use of inexpensive sensor networks to increase query ranking in some categories. Using user's preferences and selecting P(0.5) with temporal attributes it is shown by simulation to have higher accuracy, when querying accidental small fires(BA< 1 ha.). The extended study of Fire Weather Index(FWI) shows that it is invariant to temporal attributes and has higher precession in querying large fires (BA> 50 ha.), when alarm ranking counts.
