Experimental Study on Decision Fusion Parameters using Alpha Integration

This paper presents an experimental study of the convergence and results of decision fusion parameters using alpha integration algorithms. The decision fusion methods are applied to the problem of detection of ultrasound pulses inserted in high-level background noise. The implemented alpha integration algorithms were compared with the following single classifiers: support vector machine, quadratic discriminant analysis, and random forest, and with the results of the fusion using the mean of the scores provided for those single classifiers. The convergence of the parameters of alpha integration-based methods is analyzed as well as the distribution of the scores (posterior probability) of all the implemented methods. The results show the capabilities of the alpha-integration methods to improve the classification performance over the other methods considered in comparisons.