The Effect of Repeated Measurements on Bayesian Decision Regions for Class Discrimination of Time-Dependent Biological Systems

摘要

Nowadays, it is common to use nondestructive sensors to monitor property variations in biological systems. The repeated observations on the time varying system are referred to as repeated measurements. In many applications, it is important to develop a Bayesian classifier based on repeated measurements data to assure proper class identification. However, its implementation is complex due to the multidimensional and discontinuous nature of the decision boundaries. In this work, the problem of correlated data to develop a Bayesian Classifier for a multiclass problem is addressed. The effect of correlation on the classification error rate is discussed. It was found that additional correlated data does not improve the classifier likelihood for highly correlated repeated measures. Also, it is shown that error classification is adversely affected by correlation between repeated measures. Finally, a strategy to develop a multiclass Bayesian classifier from multisensory repeated measurements data is presented.