Classification of multi-source data using predictive ability measure
C.C. Chong, Ju Jia · 2002
The predictive ability of an evidence source for an uncertain event is referred to the ability of the evidence source to predict the probability of the event concerned. A new algorithm which incorporates the predictive ability of the multi-source data into the classification process is presented. The algorithm was developed based on the concept of second-order probability. Experimental results obtained show that the new algorithm outperformed the conventional multivariate Gaussian maximum likelihood classifier when applied to untrained test data. It is also shown to be more consistent in performance.