A hybrid neural knowledge expert system with parallel coordinates visualization in dengue diagnosis prediction

Abstract

This research presents the extension of DengueViz, a hybrid neural-knowledge-based expert system integrated with parallel coordinates as its visualization technique to assist in the diagnosis and severity assessment of dengue. This implementation involves the expert system with 140 rules for the classification of dengue, along with the multilayer perceptron with the stochastic gradient descent algorithm as the artificial neural network to learn data representations, support vector machine to systematically verify errors, and the Theil-Sen estimator to increase its robustness against outliers. The integration of parallel coordinates visually presents the large amount of dengue information into a single visualization space, where data interactions such as the selection of axes, filtering and highlighting reduces the clutter for it to be more comprehensible and enhances the correlation between the attributes of the information. The experiments of this system are conducted with the technology acceptance model, where the usefulness and ease of use of the system is analyzed.